EP4056898A1 - Biomass heating system with a control device optimized by means of machine learning and corresponding method - Google Patents

Abstract

Biomass heating system (1) for burning biogenic fuel, comprising: a boiler (11) with a combustion device (2) and with a heat exchanger (3); a controller (100) having a memory (105, 105a); at least one sensor (86, 111-117, 582, 592) for providing sensor data, which can detect at least one chemical and/or physical variable of the biomass heating system (1) and which communicates with the control device (100). connected is; at least one actuator (4, 5, 52, 6, 61, 66, 7, 72, 91, 201, 231) of the biomass heating system (1), which is communicatively connected to the control device (100) and can be controlled by it ; wherein the control device (100) contains at least one AI model (104) for controlling the biomass heating system (1) based on the sensor data provided, the AI model (104) being parameterized using machine learning.

Description

TECHNICAL AREA

The invention relates to a biomass heating system with a control device optimized by means of machine learning and a corresponding method.

In particular, the invention relates to a biomass heating system with optimized regulation by a control device, which has a control behavior optimized by machine learning.

STATE OF THE ART

Biomass heating systems, in particular biomass boilers, in a power range from 20 to 500 kW are known. Biomass can be considered a cheap, domestic, crisis-proof and environmentally friendly fuel. There are, for example, wood chips or pellets as combustible biomass or biogenic solid fuels.

The pellets usually consist of wood shavings, sawdust, biomass or other material that has been compacted into small discs or cylinders approximately 3 to 15 mm in diameter and 5 to 30 mm long. Wood chips (also known as wood chips, woodchips or woodchips) are wood that has been crushed with cutting tools.

Biomass heating systems for fuels in the form of pellets and wood chips essentially have a boiler with a combustion chamber (the combustion chamber) and a heat exchange device connected to it. Due to the stricter legal regulations in many countries, some biomass heating systems also have a fine dust filter. Other various accessories are regularly available, such as fuel delivery devices, control devices, probes, safety thermostats, pressure switches, flue gas or exhaust gas recirculation, boiler cleaning and a separate fuel tank.

A device for supplying fuel, a device for supplying air and an ignition device for the fuel are regularly provided in the combustion chamber. In turn, the means for supplying the air normally comprises a low-pressure fan in order to favorably influence the thermodynamic factors of combustion in the combustion chamber. A device for supplying fuel can be provided, for example, with a lateral insert (so-called transverse insert firing). The fuel is pushed into the combustion chamber from the side via a screw or a piston.

In the combustion chamber of a fixed-bed furnace, a furnace grate is also usually provided, on which the fuel is essentially supplied and burned continuously. This grate stores the fuel for combustion and has openings, for example slots, which allow the passage of part of the combustion air as primary air to the fuel. Next, the grate can be rigid or movable. There are also grate furnaces in which the combustion air is not fed through the grate but only from the side.

When the primary air flows through the grate, the grate is also cooled, which protects the material. In addition, insufficient air supply can lead to slag formation on the grate. In particular furnaces that are to be charged with different fuels, with which the present disclosure is particularly concerned, have the inherent problem that the different fuels have different ash melting points, water content and different burning behavior. It is therefore problematic to provide a heating system that is equally well suited for different fuels. The combustion chamber can also be regularly divided into a primary combustion zone (immediate combustion of the fuel on the grate and in the gas space above it before additional combustion air is supplied) and a secondary combustion zone (post-combustion zone of the flue gas after additional air supply). Drying, pyrolytic decomposition, gasification of the fuel and charcoal combustion take place in the combustion chamber. In order to completely burn the resulting combustible gases, additional combustion air is introduced in one or more stages (secondary air or tertiary air) at the start of the secondary combustion zone.

After drying, the combustion of the pellets or wood chips essentially has two phases. In the first phase, the fuel is at least partially pyrolytically decomposed and converted into gas by high temperatures and air that can be blown into the combustion chamber. In the second phase, the (partial) part that has been converted into gas is burned, as well as the burning of any remaining solids (e.g. charcoal). In this respect, the fuel outgasses, and the resulting gas and the charcoal contained therein are also burned.

Pyrolysis is the thermal decomposition of a solid substance in the absence of oxygen. Pyrolysis can be divided into primary and secondary pyrolysis. The products of primary pyrolysis are pyrolysis coke and pyrolysis gases, the pyrolysis gases being divided into room temperature condensable and non-condensable gases. The primary pyrolysis takes place at roughly 250-450°C and the secondary pyrolysis at around 450-600°C. The secondary pyrolysis that subsequently occurs is based on the further reaction of the pyrolysis products that were primarily formed. The drying and pyrolysis take place at least largely without the use of air, since volatile CH compounds escape from the particle and therefore no air can reach the particle surface. Gasification can be seen as part of oxidation; the solid, resulting from the pyrolytic decomposition, liquid and gaseous products are reacted by further exposure to heat. This is done by adding a gasification agent such as air, oxygen, water vapor or carbon dioxide. The lambda value during gasification is greater than zero and less than one. Gasification takes place at around 300 to 850°C or even up to 1,200°C. The complete oxidation with excess air (lambda greater than 1) takes place by adding more air to these processes. The end products of the reaction are essentially carbon dioxide, water vapor and ash. In all phases, the boundaries are not rigid, but fluid. The combustion process can be advantageously regulated by means of a lambda probe provided at the exhaust gas outlet of the boiler.

Generally speaking, converting the pellets into gas increases combustion efficiency because gaseous fuel is better mixed with the combustion air and thus more completely converted, and lower emissions of pollutants, unburned particles and ash (fly ash or dust particles) are produced .

The combustion of biomass produces gaseous or airborne combustion products, the main components of which are carbon, hydrogen and oxygen. These can be divided into emissions from complete oxidation, from incomplete oxidation and substances from trace elements or impurities. The emissions from complete oxidation are essentially carbon dioxide (CO ₂ ) and water vapor (H ₂ O). The formation of carbon dioxide from the carbon in the biomass is the goal of combustion, as the released energy can be used more fully. The release of carbon dioxide (CO ₂ ) is largely proportional to the carbon content of the fuel burned; thus the carbon dioxide is also dependent on the useful energy to be provided. A reduction can essentially only be achieved by improving the efficiency. Combustion residues, such as ash or slag, are also produced.

However, the complex combustion processes described above are not easy to control. Generally speaking, there is a need for improvement with regard to the combustion processes in biomass heating systems.

In addition to the air supply to the combustion chamber, flue gas or exhaust gas recirculation devices are also known, which recirculate exhaust gas from the boiler to the combustion chamber for cooling and renewed combustion. In the prior art, there are usually openings in the combustion chamber for supplying primary air through a primary air line feeding the combustion chamber, and there are also peripheral openings in the combustion chamber for supplying secondary air from a secondary air line or possibly fresh air. Flue gas recirculation can take place below or above the grate. In addition, the flue gas can be recirculated mixed with the combustion air or separately.

The flue gas or exhaust gas from combustion in the combustion chamber is fed to the heat exchanger so that the hot combustion gases flow through the heat exchanger to transfer heat to a heat exchange medium, which is normally water at around 80°C (usually between 70 and 70°C). °C and 110°C). The boiler further usually has a radiant part, which is integrated into the combustion chamber, and a convection part (the heat exchanger connected thereto).

The ignition device is mostly a hot air device or a glow device. In the first case, combustion is started by supplying hot air to the combustion chamber, the hot air being heated by an electrical resistance. In the second case, the ignition device comprises a glow plug/rod or several glow plugs to heat the pellets or wood chips by direct contact until combustion begins. The glow plugs can also be equipped with a motor to remain in contact with the pellets or wood chips during the ignition phase and then move back to avoid being exposed to the flames. This solution is subject to wear and tear and expensive.

There are also a variety of applications, such as automated diagnostic methods, detection of credit card fraud, stock market analysis, classification of nucleotide sequences, speech and text recognition, and autonomous systems, devices, and methods that use so-called "machine learning". An artificial intelligence learns from examples, such as sensor data, and can apply this in a generalized way once the learning phase has ended. For this purpose, machine learning algorithms build a (statistical) model (in this case an AI model of a biomass heating system) that is based on training data. In this case, the model can carry out a calculation using a neural network, which is parameterized by training in its weightings or weightings.

In essence, with this system, the examples are not simply learned by heart, but patterns and regularities in the learning data are recognized in order to parameterize an AI model. The weights of the neural network can be processed and/or stored collectively as parameter data. An example of such parameter data (weights of a neural network) is in 24 shown.

Artificial intelligence (AI) is a sub-area of computer science. It mimics human cognitive abilities by recognizing and sorting information from input data. This intelligence can be based on programmed processes or generated by machine learning. A distinction is usually made between supervised learning and non-supervised learning.

In supervised learning, the algorithm learns a function from given pairs of inputs and outputs. A "teacher" provides the correct function value for an input during the learning process. The aim of supervised learning is that after several calculations with different inputs and outputs, the neural network is trained to create associations. A sub-area of supervised learning is automatic classification.

In the case of unsupervised or unsupervised learning, the algorithm generates a statistical model for a given set of inputs that describes the inputs and contains recognized categories and relationships, thus enabling predictions. There are clustering methods that divide the (raw) data into several categories that differ from one another in characteristic patterns. The network thus independently creates classifiers according to which it divides the input pattern. An important algorithm in this context is the EM algorithm, which iteratively sets the parameters of a model in such a way that it optimally explains the data seen. He assumes the existence of unobservable categories and alternately estimates the membership of the data in one of the categories and the parameters that make up the categories. One application of the EM algorithm can be found, for example, in the Hidden Markov Models (HMMs). Other methods of unsupervised learning, e.g. B. principal component analysis, dispense with the categorization. They aim to translate the observed data into a simpler representation that reproduces them as accurately as possible despite drastically reduced information.

The so-called "deep learning" refers to a method of machine learning that uses artificial neural networks with numerous intermediate layers (English: hidden layers) between the input layer and the output layer and thus an extensive inner structure emerges. It is a special method of information processing known to the person skilled in the art, the understanding of which is a prerequisite for this application.

Deep learning, a special type of machine learning that is commonly understood, uses a series of hierarchical layers, or hierarchy of concepts, to perform the machine learning process. The artificial neural networks used here contain so-called neurons, which are connected to one another like a network. The first layer of the neural network, the visible input layer, processes a raw data input, for example sensor data, such as the individual temperature values from a temperature sensor. the Data input contains variables that are accessible to observation, hence this is called the "visible layer".

This first layer forwards its outputs to the next layer. This second layer processes the information from the previous layer and also passes the result on. The next layer receives the information from the second layer and processes it further. These layers are called hidden layers. The features they contain are becoming increasingly abstract. Their values are not given in the original data. Instead, the model must determine which concepts are useful in explaining the relationships in the observed data. This continues through all layers of the artificial neural network. The result is output in the visible last layer. This breaks the desired complex data processing into a series of nested simple mappings, each described by a different layer of the model.

In summary, neural networks are machine learning models that employ one or more layers of nonlinear units to predict an output for a received input. A neural network that has been trained or taught can be used as an AI model (to be trained and trained) for the biomass heating system.

Some neural networks are deep neural networks that include one or more hidden layers in addition to an output view. The output of each hidden layer is used as input to the next layer in the network, i.e. H. the next hidden layer or the output layer. Each layer of the network produces an output from a received input in accordance with current values of a corresponding set of parameters. The trained neural network is then the core of what is generally referred to as "artificial intelligence".

A deep neural network or so-called "deep learning" can be implemented with the "TensorFlow" ^™ software. TensorFlow ^™ is a framework for stream-oriented programming. TensorFlow ^™ is popular in the field of machine learning. The name "TensorFlow" ^™ comes from arithmetic operations performed by artificial neural networks on multidimensional data fields, so-called tensors. TensorFlow ^™ was originally developed by the Google Brain team for Google's internal needs and released in 2015 under the Apache 2.0 open source license. With this framework, what is presented later in the exemplary embodiments can be implemented, which means that the person skilled in the art can also derive his understanding of an AI model.

Machine learning (ML) technology is often understood as a subfield of artificial intelligence. In this respect, the present application also relates to the novel use of artificial intelligence for improved control of a biomass heating system.

The basic problems with conventional biomass heating systems are that the gaseous or solid emissions are too high, that the efficiency is too low and that the dust emissions are too high. Another problem is the varying quality of the fuel, due to the varying water content and the lumpiness of the fuel, which makes it difficult to burn the fuel evenly with low emissions. In particular in the case of biomass heating systems, which are intended to be suitable for different types of biological or biogenic fuel, the varying quality and consistency of the fuel makes it difficult to maintain consistently high efficiency of the biomass heating system.

These problems often lead to the user of the biomass heating system having to intervene manually and, for example, to make control parameters or basic settings himself in order to optimize or even enable the operation of the system. The biomass heating system is often adjusted according to "gut feeling" or after many years of manual recording and optimization attempts, which makes the operation of the biomass heating system more difficult and disadvantageously complicated.

In addition, the controls of conventional biomass heating systems are regularly quite primitive and have an inadequate control behavior. According to the inventor's experience, there are often excessive control fluctuations, undesirable control reactions and/or a greatly delayed control behavior in conventional biomass heating systems.

For example, the control of valves in recirculation equipment often uses only a primitive step value algorithm based on a single sensed value (usually the boiler temperature), so the air recirculation is subject to regular fluctuations (i.e. the corresponding control loop oscillates). This ultimately makes stationary and optimized control of the boiler output impossible. The same applies to many other control tasks in conventional biomass heating systems (power, lambda, etc.) for which there is a need for optimization.

As a result, the conventional regulations often lead to efficiency losses in the biomass heating systems, increased pollutant emissions and increased maintenance costs.

There is therefore a considerable need for optimization in this regard.

Biomass heating systems for pellets or wood chips have the following additional disadvantages and problems.

It can be an object of the invention to provide a biomass heating system in hybrid technology, which is low in emissions (especially with regard to fine dust, CO, hydrocarbons, NOx) and which can be operated easily.

In addition, the biomass heating system should be able to be operated in a fuel-flexible manner with wood chips and pellets, and at the same time have a high level of efficiency.

According to the invention and in addition, the following considerations can play a role:
The hybrid technology should enable the use of both pellets and wood chips with a water content of between 8 and 35 percent by weight.

Gaseous emissions that are as low as possible (less than 50 or 100 mg/Nm ³ based on dry flue gas and 13 percent by volume O ₂ ) should be achieved.

Very low dust emissions of less than 15 mg/Nm ³ without and less than 5 mg/Nm ³ with electrostatic precipitator operation are aimed for.

A high degree of efficiency of up to 98% (based on the fuel energy supplied (calorific value) should be achieved.

Furthermore, one can take into account that the operation of the system should be optimized. For example, simple ash removal, simple cleaning or simple maintenance should be made possible.

In addition, there should be high system availability.

The task(s) mentioned above or the potential individual problems can also relate to individual partial aspects of the overall system, for example to the regulation and/or control.

The above object(s) is/are solved by the subject matter of the independent claims. Further aspects and advantageous developments are the subject matter of the dependent claims.

• einen Kessel mit einer Brenneinrichtung und mit einem Wärmetauscher ;
• eine Steuereinrichtung mit einem Speicher ;
• zumindest einen Sensor zur Bereitstellung von Sensordaten, welcher zumindest eine chemische und/oder physikalische Größe der Biomasse-Heizanlage erfassen kann und welcher mit der Steuereinrichtung kommunikativ verbunden ist;
• zumindest einen Aktor der Biomasse-Heizanlage, welcher mit der Steuereinrichtung kommunikativ verbunden ist und von dieser angesteuert werden kann;
• wobei die Steuereinrichtung zumindest ein KI-Modell zur Regelung der Biomasse-Heizanlage basierend auf den bereitgestellten Sensordaten beinhaltet, wobei das KI-Modell mittels maschinellem Lernen parametriert wird.

According to one aspect, a biomass heating system, preferably for firing solid fuel (e.g. in the form of pellets and/or wood chips) or also for firing biogenic (gaseous) fuel), is disclosed, the biomass heating system having the following:

• a boiler with a burner and with a heat exchanger;
• a control device with a memory;
• at least one sensor for providing sensor data, which can detect at least one chemical and/or physical variable of the biomass heating system and which is communicatively connected to the control device;
• at least one actuator of the biomass heating system, which is communicatively connected to the control device and can be controlled by it;
• wherein the control device contains at least one AI model for controlling the biomass heating system based on the sensor data provided, the AI model being parameterized using machine learning.

Biomass heating system according to the preceding aspect, wherein
the AI model is a neural network with at least 3 layers and at least 200 neurons.

• die Steuereinrichtung derart eingerichtet ist, dass
• die Sensordaten über einen vorbestimmten Zeitraum während eines Betriebs der Biomasse-Heizanlage aggregiert werden, und
• die aggregierten Sensordaten zum maschinellen Lernen verwendet werden.

Biomass heating system according to one of the preceding aspects, wherein

• the control device is set up in such a way that
• the sensor data are aggregated over a predetermined period of time during operation of the biomass heating system, and
• the aggregated sensor data is used for machine learning.

• die Steuereinrichtung der Biomasse-Heizanlage eine maschinelle Lerneinheit aufweist; und
• die Biomasse-Heizanlage derart eingerichtet ist, dass
• die von dem zumindest einem Sensor erfasste Größe als die Sensordaten in dem Speicher der Steuereinrichtung abgespeichert werden; und
• das maschinelle Lernen mittels der maschinellen Lerneinheit unter Verwendung der in dem Speicher abgespeicherten Sensordaten erfolgt.

Biomass heating system according to one of the preceding aspects, wherein

• the controller of the biomass heating system has a machine learning unit; and
• the biomass heating system is set up in such a way that
• the variable detected by the at least one sensor is stored as the sensor data in the memory of the control device; and
• the machine learning takes place by means of the machine learning unit using the sensor data stored in the memory.

• die Biomasse-Heizanlage derart eingerichtet ist, dass
• die von dem zumindest einem Sensor erfasste zumindest eine Größe als die Sensordaten in dem Speicher der Steuereinrichtung abgespeichert wird; und
• die in dem Speicher der Steuereinrichtung gespeicherten Sensordaten über eine Netzwerkverbindung an eine zentrale Recheneinheit übertragen werden; und
• von der zentralen Recheneinheit ein KI-Modell oder Gewichtsdaten zur Regelung der Biomasse-Heizanlage an die Steuereinrichtung der Biomasse-Heizanlage übertragen wird/werden, wobei das KI-Modell durch maschinelles Lernen mittels einer maschinellen Lerneinheit der zentralen Recheneinheit zumindest unter Verwendung der übertragenen Sensordaten parametriert wird.

Biomass heating system according to one of the preceding aspects, wherein

• the biomass heating system is set up in such a way that
• the at least one variable detected by the at least one sensor is stored as the sensor data in the memory of the control device; and
• the sensor data stored in the memory of the control device are transmitted to a central processing unit via a network connection; and
• an AI model or weight data for controlling the biomass heating system is/are transmitted from the central processing unit to the control device of the biomass heating system, the AI model being parameterized by machine learning by means of a machine learning unit of the central processing unit, at least using the transmitted sensor data becomes.

• das KI-Modell eingerichtet ist, eine Klassifizierung des Brennstoffs durchzuführen, und
• die Klassifizierung des Brennstoffs auf folgenden Sensordaten basiert:
  einem Sauerstoffgehalt des Abgases der Biomasse-Heizanlage, welcher mit einer Lambda-Sonde erfasst wird; sowie zumindest eines des Folgenden:
  • einer Abgastemperatur des Abgases der Biomasse-Heizanlage, welche mit einem Abgastemperatursensor erfasst wird;
  • einer Brennkammertemperatur eines Brennraums der Biomasse-Heizanlage, welche mit einem Brennraumtemperatursensor erfasst wird;
  • wobei der Sauerstoffgehalt, die Abgastemperatur und die Brennkammertemperatur in einem Zeitraum nach einer Zündung des Brennstoffs durch eine Zündeinrichtung bis zum Erreichen eines vordefinierten Sauerstoffgehalts, vorzugsweise 16 Vol %-O2, erfasst werden.

Biomass heating system according to one of the preceding aspects, wherein

• the AI model is set up to classify the fuel, and
• the classification of the fuel is based on the following sensor data:
  an oxygen content of the exhaust gas of the biomass heating system, which is detected with a lambda probe; and at least one of the following:
  • an exhaust gas temperature of the exhaust gas of the biomass heating system, which is detected with an exhaust gas temperature sensor;
  • a combustion chamber temperature of a combustion chamber of the biomass heating system, which is detected with a combustion chamber temperature sensor;
  • wherein the oxygen content, the exhaust gas temperature and the combustion chamber temperature are recorded over a period of time after the fuel has been ignited by an ignition device until a predefined oxygen content, preferably 16% by volume O2, has been reached.

• die Steuereinrichtung mit dem KI-Modell eingerichtet ist, eine Regelung eines Motors einer Brennstoffzufuhr in die Brenneinrichtung auf eine vorgegebene Brennstoffbetthöhe durchzuführen, wobei die Regelung auf zumindest einem der folgenden Sensordaten basiert:
  einem Sauerstoffgehalt des Abgases der Biomasse-Heizanlage, welcher mit einer Lambda-Sonde erfasst wird; sowie zumindest eines des Folgenden:
  • einer Abgastemperatur des Abgases der Biomasse-Heizanlage, welche mit einem Abgastemperatursensor erfasst wird;
  • einer Brennkammertemperatur eines Brennraums der Biomasse-Heizanlage, welche mit einem Brennraumtemperatursensor erfasst wird.
  • Biomasse-Heizanlage gemäß einem der vorhergehenden Aspekte, wobei
• die Steuereinrichtung mit dem KI-Modell eingerichtet ist, eine Lambda-Regelung der Biomasse-Heizanlage auf einen vorbestimmten Lambda-Wert des Abgases durchzuführen, wobei die Regelung auf zumindest einem der folgenden Sensordaten basiert:
  einem Sauerstoffgehalt des Abgases der Biomasse-Heizanlage, welcher mit einer Lambda-Sonde erfasst wird; sowie zumindest eines des Folgenden:
  • einer Abgastemperatur des Abgases der Biomasse-Heizanlage, welche mit einem Abgastemperatursensor erfasst wird;
  • einer Brennkammertemperatur einer Brennkammer der Biomasse-Heizanlage, welche mit einem Brennraumtemperatursensor erfasst wird;
  • einem Unterdruck in der Brennkammer, welcher durch einen Unterdrucksensor erfasst wird;
  • einer Brennstoffbetthöhe auf dem Drehrost , welche durch einen Brennstoffbetthöhensensor erfasst wird;
  • und wobei durch das KI-Modell zumindest eines der Luftventile der Rezirkulationseinrichtung angesteuert wird.

Biomass heating system according to one of the preceding aspects, wherein

• the control device is set up with the AI model to regulate a motor of a fuel feed into the combustion device to a specified fuel bed level, the regulation being based on at least one of the following sensor data:
  an oxygen content of the exhaust gas of the biomass heating system, which is detected with a lambda probe; and at least one of the following:
  • an exhaust gas temperature of the exhaust gas of the biomass heating system, which is detected with an exhaust gas temperature sensor;
  • a combustion chamber temperature of a combustion chamber of the biomass heating system, which is detected with a combustion chamber temperature sensor.
  • Biomass heating system according to one of the preceding aspects, wherein
• the control device is set up with the AI model to carry out lambda control of the biomass heating system to a predetermined lambda value of the exhaust gas, the control being based on at least one of the following sensor data:
  an oxygen content of the exhaust gas of the biomass heating system, which is detected with a lambda probe; and at least one of the following:
  • an exhaust gas temperature of the exhaust gas of the biomass heating system, which is detected with an exhaust gas temperature sensor;
  • a combustion chamber temperature of a combustion chamber of the biomass heating system, which is detected with a combustion chamber temperature sensor;
  • a negative pressure in the combustion chamber, which is detected by a negative pressure sensor;
  • a fuel bed level on the rotary grate 12 sensed by a fuel bed level sensor;
  • and wherein at least one of the air valves of the recirculation device is controlled by the AI model.

• eine Brennkammertemperatur einer Brennkammer der Biomasse-Heizanlage, welche mit einem Brennraumtemperatursensor erfasst wird;
• einer Brennstoffbetthöhe auf dem Drehrost , welche durch einen Brennstoffbetthöhensensor erfasst wird;
• eine Kesseltemperatur, welche eine Temperatur des Wärmetauschmediums im Kessel ist;
• eine abgegebene Wärmemenge des Kessels, welche mittels eines Wärmemengensensors erfasst wird;
• Wetterprognosedaten, welche über ein Netzwerk, vorzugsweise das Internet, abgerufen werden; und
• und wobei durch das KI-Modell zumindest eines der Luftventile der Rezirkulationseinrichtung und/oder der Primärluftzufuhr angesteuert wird, oder das Gebläse der Biomasse-Heizanlage angesteuert wird.

Biomass heating system according to one of the preceding aspects, wherein
the control device is set up with the AI model to regulate the output of the biomass heating system to a predetermined output value of the biomass heating system (1), the control being based on at least one of the following sensor data and/or external data:

• a combustion chamber temperature of a combustion chamber of the biomass heating system, which is detected with a combustion chamber temperature sensor;
• a fuel bed level on the rotary grate 12 sensed by a fuel bed level sensor;
• a boiler temperature, which is a temperature of the heat exchange medium in the boiler;
• an amount of heat emitted by the boiler, which is detected by means of an amount of heat sensor;
• Weather forecast data retrieved via a network, preferably the Internet; and
• and wherein at least one of the air valves of the recirculation device and/or the primary air supply is controlled by the AI model, or the blower of the biomass heating system is controlled.

• einer Brennkammertemperatur einer Brennkammer der Biomasse-Heizanlage, welche mit einem Brennraumtemperatursensor erfasst wird;
• einer Brennstoffbetthöhe auf dem Drehrost , welche durch einen Brennstoffbetthöhensensor erfasst wird;
• einem Sauerstoffgehalt des Abgases der Biomasse-Heizanlage, welcher mit einer Lambda-Sonde erfasst wird;
• eine Kesseltemperatur, welche eine Temperatur des Wärmetauschmediums im Kessel ist;
• eine abgegebene Wärmemenge des Kessels, welche mittels eines Wärmemengensensors erfasst wird;

Biomass heating system according to one of the preceding aspects, wherein
the control device is set up with the AI model to regulate the position of a valve of a primary air duct of the biomass heating system (1) to a predetermined amount of primary air, the regulation being based on at least one of the following sensor data:

• a combustion chamber temperature of a combustion chamber of the biomass heating system, which is detected with a combustion chamber temperature sensor;
• a fuel bed level on the rotary grate 12 sensed by a fuel bed level sensor;
• an oxygen content of the exhaust gas of the biomass heating system, which is detected with a lambda probe;
• a boiler temperature, which is a temperature of the heat exchange medium in the boiler;
• an amount of heat emitted by the boiler, which is detected by means of an amount of heat sensor;

• die Steuereinrichtung mit dem KI-Modell eingerichtet ist, eine Ein/Aus-Regelung der Biomasse-Heizanlage durchzuführen, wobei die Regelung auf Folgendem basiert:
  • Wetterprognosedaten, welche über ein Netzwerk, vorzugsweise das Internet, abgerufen werden; und
  • eine vorgegebene Kesselsolltemperatur;
  • eine Kesseltemperatur, welche eine Temperatur des Wärmetauschmediums im Kessel ist;
• sowie optional auf zumindest einem der folgenden Sensordaten basieren kann:
  • eine Temperatur eines Wärmespeichers, vorzugsweise eines Puffers, in einem Heizkreislauf, welcher von der Biomasse-Heizanlage mit Energie versorgt wird;
  • eine Leistungsabgabefähigkeit der Biomasse-Heizanlage, beispielsweise die Leistungsklasse des Kessels (kW);
  • eine abgegebene Wärmemenge des Kessels, welche mittels eines Wärmemengensensors erfasst wird.

Biomass heating system according to one of the preceding aspects, wherein

• the controller with the AI model is set up to perform on/off control of the biomass heating system, the control being based on:
  • Weather forecast data retrieved via a network, preferably the Internet; and
  • a predetermined boiler setpoint temperature;
  • a boiler temperature, which is a temperature of the heat exchange medium in the boiler;
• and optionally based on at least one of the following sensor data:
  • a temperature of a heat accumulator, preferably a buffer, in a heating circuit, which is supplied with energy by the biomass heating system;
  • a power output capability of the biomass heating system, for example the power class of the boiler (kW);
  • an amount of heat given off by the boiler, which is recorded by means of a heat quantity sensor.

Biomass heating system according to one of the preceding aspects, wherein
the result of the calculation of the trained AI model being used as at least one input variable for a control algorithm, for example a P, PI, PID or PD control.

• einen Kessel mit einer Brenneinrichtung und einem Wärmetauscher ;
• eine Steuereinrichtung mit einem Speicher ;
• wobei das Verfahren das Folgende aufweist:
  Regeln der Biomasse-Heizanlage mittels eines KI-Modells der Steuereinrichtung, welches durch maschinelles Lernen parametriert wird.

Process for controlling a biomass heating system for firing solid fuel in the form of pellets and/or wood chips, the biomass heating system having:

• a boiler with a burner and a heat exchanger;
• a control device with a memory;
• the method comprising:
  Regulation of the biomass heating system using an AI model of the control device, which is parameterized by machine learning.

• Erfassen von zumindest einer chemischen und/oder physikalischen Größe der Biomasse-Heizanlage durch zumindest einen Sensor der Biomasse-Heizanlage , welcher mit der Steuereinrichtung kommunikativ verbunden ist; und
• Verarbeiten der zumindest einer chemischen und/oder physikalischen Größe unter Verwendung des KI-Modells; und
• Ansteuern des zumindest eines Aktors der Biomasse-Heizanlage, welcher mit der Steuereinrichtung kommunikativ verbunden ist, basierend auf dem Ergebnis des Schritts des Verarbeitens.

Method according to the preceding aspect, wherein
controlling the biomass heating system includes:

• Detection of at least one chemical and/or physical variable of the biomass heating system by at least one sensor of the biomass heating system, which is communicatively connected to the control device; and
• processing the at least one chemical and/or physical variable using the AI model; and
• Controlling the at least one actuator of the biomass heating system, which is communicatively connected to the control device, based on the result of the processing step.

• Aggregieren der Sensordaten über einen vorbestimmten Zeitraum während eines Betriebs der Biomasse-Heizanlage, und
• Verwenden der aggregierten Sensordaten zum maschinellen Lernen.

Method according to one of the preceding aspects, the method further comprising the following steps:

• Aggregating the sensor data over a predetermined period of time during operation of the biomass heating system, and
• Using the aggregated sensor data for machine learning.

• die Steuereinrichtung der Biomasse-Heizanlage eine maschinelle Lerneinheit aufweist; und
• das Verfahren weiter den folgenden Schritt aufweist:
  Abspeichern der von der Mehrzahl von Sensoren erfassten zumindest einen Größe in dem Speicher der Steuereinrichtung als die Sensordaten; wobei das maschinelle Lernen mittels der maschinellen Lerneinheit unter Verwendung der in dem Speicher gespeicherten Sensordaten erfolgt.

Method according to one of the preceding aspects, wherein

• the controller of the biomass heating system has a machine learning unit; and
• the method further comprises the following step:
  storing the at least one variable detected by the plurality of sensors in the memory of the control device as the sensor data; wherein the machine learning is performed by the machine learning unit using the sensor data stored in the memory.

• Speichern der von dem zumindest einem Sensor erfassten zumindest einen Größe in dem Speicher der Steuereinrichtung als die Sensordaten; und
• Übertragen der in dem Speicher der Steuereinrichtung gespeicherten Sensordaten über eine Netzwerkverbindung an eine zentrale Recheneinheit; und
• Parametrieren des KI-Modell durch maschinelles Lernen mittels einer maschinellen Lerneinheit der zentralen Recheneinheit unter Verwendung der übertragenen Sensordaten; und
• Übertragen des parametrisierten KI-Modells an die Steuereinrichtung der Biomasse-Heizanlage von der zentralen Recheneinheit .

Method according to one of the preceding aspects, the method further comprising the following steps:

• storing the at least one variable detected by the at least one sensor in the memory of the control device as the sensor data; and
• Transmission of the sensor data stored in the memory of the control device to a central processing unit via a network connection; and
• Parameterizing the AI model through machine learning using a machine learning unit of the central processing unit using the transmitted sensor data; and
• Transmission of the parameterized AI model to the control device of the biomass heating system from the central processing unit.

Method for controlling a biomass heating system for firing solid fuel in the form of pellets and/or wood chips, comprising the biomass heating system and the processes according to one of the preceding aspects.

• einen Kessel mit einer Brenneinrichtung und einem Wärmetauscher ;
• eine Steuereinrichtung mit einem Speicher ;
• eine Mehrzahl von Sensoren zur Bereitstellung von Sensordaten, welche chemische und/oder physikalische Größen der Biomasse-Heizanlage erfassen können und welche mit der Steuereinrichtung kommunikativ verbunden sind;
• eine Mehrzahl von Aktoren der Biomasse-Heizanlage, welche mit der Steuereinrichtung kommunikativ verbunden sind und von dieser angesteuert werden können;
• wobei das Verfahren die folgenden Schritte aufweist:
  • Erfassen von zumindest einer chemischen und/oder physikalischen Größe der Biomasse-Heizanlage durch zumindest einen Sensor (86, 111-117, 582), 592 der Biomasse-Heizanlagen; und
  • Abspeichern der erfassten Größen in den Speichern der Steuereinrichtungen der Biomasse-Heizanlagen als Sensordaten;
  • Übertragen zumindest eines Teils der in den Speichern der Steuereinrichtungen der Biomasse-Heizanlagen gespeicherten Größen über zumindest eine Netzwerkverbindung an die zentrale Recheneinheit.

Method for a system with a plurality of biomass heating systems and with at least one central computing device, the biomass heating systems being communicatively connected to the central computing device via a network, the plurality of biomass heating systems having the following:

• a boiler with a burner and a heat exchanger;
• a control device with a memory;
• a plurality of sensors for providing sensor data, which can detect chemical and/or physical quantities of the biomass heating system and which are communicatively connected to the control device;
• a plurality of actuators of the biomass heating system, which are communicatively connected to the control device and can be controlled by it;
• the method comprising the following steps:
  • Detection of at least one chemical and/or physical variable of the biomass heating system by at least one sensor (86, 111-117, 582), 592 of the biomass heating system; and
  • Saving the recorded variables in the memories of the control devices of the biomass heating systems as sensor data;
  • Transmission of at least some of the variables stored in the memories of the control devices of the biomass heating systems via at least one network connection to the central processing unit.

• Aggregieren der übertragenen Sensordaten in der zentralen Recheneinheit;
• Maschinelles Lernen mittels der maschinellen Lerneinheit unter Verwendung der aggregierten Sensordaten, woraus ein parametriertes KI-Modell resultiert;
• Übertragen des parametrierten KI-Modells an die Steuereinrichtung der Biomasse-Heizanlage von der zentralen Recheneinheit über das Netzwerk;
• Regeln der Biomasse-Heizanlage mittels des übertragenen KI-Modells durch die der Steuereinrichtung.
• Verfahren gemäß einem der vorhergehenden Aspekte, weiter aufweisend die Merkmale von zumindest von einem der vorstehenden vorrichtungsbezogenen Aspekte.

Method according to one of the preceding aspects, further comprising the following steps:

• Aggregating the transmitted sensor data in the central processing unit;
• Machine learning using the machine learning unit using the aggregated sensor data, resulting in a parameterized AI model;
• Transmission of the parameterized AI model to the control device of the biomass heating system from the central processing unit via the network;
• Regulation of the biomass heating system using the transmitted AI model by the control device.
• Method according to one of the preceding aspects, further comprising the features of at least one of the preceding device-related aspects.

A computer-readable storage medium containing instructions that, when executed by a computing device, cause the computing device to perform any of the foregoing methods or the operations of any of the foregoing aspects.

The advantages of this main aspect, its configuration and also the associated (individual) aspects and also the advantages of the respective methods result from the following description of the associated exemplary embodiments.

In the present case, "horizontal" can denote a level orientation of an axis or a cross section, assuming that the boiler is also set up horizontally, with which, for example, the ground level can be the reference. Alternatively, "horizontal" as used herein means "parallel" to the base plane of the vessel, as commonly defined. Further alternatively, in particular if there is no reference plane, "horizontal" can be understood merely as "parallel" to the combustion plane of the grate.

Although all of the above individual features and details of an aspect of the invention and the developments of this aspect are described in connection with the biomass heating system and the recirculation device, these individual features and details are also disclosed as such independently of the biomass heating system.

In particular, independently of the biomass heating system, an AI model-based control for a biomass heating system, the control process of the 18 ff. for a biomass heating system and an optimized self-learning function of the 22 for a biomass heating system and can be claimed independently of this. An AI model for a biomass heating system is also disclosed, which can be claimed separately, for example as a control device with an AI model for a biomass heating system. The features of all aspects disclosed above, all embodiments and modifications and those of the supplementary examples with the control method, with the self-learning function, and with the control device can be combined with an AI model for a biomass heating system, like this considered feasible by those skilled in the art.

It goes without saying that an AI model has better control behavior the more input parameters there are for learning and for control. In this respect, the "at least one of the following sensor data" described above for the aspects is an indication of a minimum parameter, with the selection of more than one parameter regularly leading to better behavior of the AI model.

Fig. 1

zeigt eine dreidimensionale Überblicksansicht einer Biomasse-Heizanlage gemäß einer Ausführungsform der Erfindung;

Fig. 2

zeigt eine Querschnittsansicht durch die Biomasse-Heizanlage der Fig. 1, welche entlang einer Schnittlinie SL1 vorgenommen wurde und welche aus der Seitenansicht S betrachtet dargestellt ist;

Fig. 3

zeigt ebenso eine Querschnittsansicht durch die Biomasse-Heizanlage der Fig. 1 mit einer Darstellung des Strömungsverlaufs, wobei die Querschnittsansicht entlang einer Schnittlinie SL1 vorgenommen wurde und aus der Seitenansicht S betrachtet dargestellt ist;

Fig. 4

zeigt eine Teilansicht der Fig. 2, die eine Brennkammergeometrie des Kessels der Fig. 2 und Fig. 3 darstellt;

Fig. 5

zeigt eine Schnittansicht durch den Kessel bzw. die Brennkammer des Kessels entlang der Vertikalschnittlinie A2 der Fig. 4;

Fig. 6

zeigt eine dreidimensionale Schnittansicht auf die Primärverbrennungszone der Brennkammer mit dem Drehrost der Fig. 4;

Fig. 7

zeigt entsprechend zur Fig. 6 eine Explosionsdarstellung der Brennkammersteine;

Fig. 8

zeigt eine Aufsicht auf den Drehrost mit Drehrostelementen von oben aus Sicht der Schnittlinie A1 der Fig. 2;

Fig. 9

zeigt den Drehrost der Fig. 2 in geschlossener Position, wobei alle Drehrostelemente horizontal ausgerichtet bzw. geschlossen sind;

Fig. 10

zeigt den Drehrost der Fig. 9 in dem Zustand einer Teilabreinigung des Drehrosts im Gluterhaltungsbetrieb;

Fig. 11

zeigt den Drehrost der Fig. 9 im Zustand der Universalabreinigung, welche bevorzugt während eines Anlagenstillstands durchgeführt wird;

Fig. 12

zeigt eine herausgestellte Schrägansicht einer beispielhaften Rezirkulationseinrichtung mit Brennkammersteinen, die eine Primärverbrennungszone umgeben;

Fig. 13

zeigt eine herausgestellte semitransparente Schrägansicht der Rezirkulationseinrichtung der Fig. 12;

Fig. 14

zeigt eine Seitenansicht der Rezirkulationseinrichtung 5 der Figuren 12 und 13;

Fig. 15

zeigt ein schematisches Blockdiagramm, dass den Strömungsverlauf in den jeweiligen Einzelkomponenten der Biomasse-Heizanlage und der Rezirkulationseinrichtung der Fig. 12 bis 14 zeigt;

Fig. 16

zeigt eine herausgestellte semitransparente Schrägansicht einer Rezirkulationseinrichtung einer weiteren Ausführungsform;

Fig. 17

zeigt ein schematisches Blockdiagramm, dass den Strömungsverlauf in den jeweiligen Einzelkomponenten einer Biomasse-Heizanlage und der Rezirkulationseinrichtung der Fig. 16 gemäß einer weiteren Ausführungsform offenbart.

Fig. 18

zeigt ein schematisches Blockdiagramm einer Vorrichtung und eines Verfahrens mit beispielhaften Komponenten der Biomasse-Heizanlage der Fig. 1 bis Fig. 17, wobei dies auf einem lokalen Ansatz des maschinellen Lernens basiert;

Figuren 19a, 19b und 19c

zeigen schematische Blockdiagramme von Vorrichtungen und Verfahren mit beispielhaften Komponenten der Biomasse-Heizanlage der Fig. 1 bis Fig. 17, wobei dies auf einem zentralen Ansatz des maschinellen Lernens basiert;

Fig. 20

zeigt ein schematisches Diagramm, das ein Grundkonzept für ein Neuron darstellt, welches eine mögliche Grundlage eines maschinellen Lernens ist;

Fig. 21

zeigt ein schematisches Diagramm, das ein dreischichtiges neuronales Netzwerk darstellt, das durch Zusammenfassen von in Fig. 21 dargestellten Neuronen gebildet wird;

Fig. 22

zeigt ein Blockdiagramm, welches einen Datenfluss bzw. einen Ablauf eines maschinellen Lernens einer Biomasse-Heizanlage und der Anwendung des erlernten Modells auf die Biomasse-Heizanlage darstellt;

Figuren 23a und 23b

zeigen Beispiele von Lernkurven beim Lernen / Trainieren des Modells

Fig. 24

zeigt beispielhafte Gewichtsdaten, welche Gewichte eines neuronalen Netzwerks beinhaltet.

Fig. 25

zeigt einen Verbrennungsbetrieb bzw. ein Betriebsverfahren der Biomasse-Heizanlage der Fig. 1 bis 24;

Fig. 26

zeigt ein schematisches Diagramm einer Anwendung des maschinellen Lernens zur Klassifizierung von Brennstoff mit diversen Eingangsparametern bzw. Sensordaten;

Figuren 27a, 27b und 27c

zeigen entsprechend Sensordaten von Sensoren, welche für die Anwendung der Fig. 26 verwendet werden;

Fig. 28

zeigt ein Verfahren zur Materialerkennung/Klassifizierung von Brennstoff der Biomasse-Heizanlage der Fig. 1 bis 24 mit einem KI-Modell;

Fig. 29

zeigt ein Verfahren zur Brennstoffbetthöhenregelung der BiomasseHeizanlage der Fig. 1 bis 24 mit einem KI-Modell;

Fig. 30

zeigt ein Verfahren zur Lambda-Regelung bzw. zur Sauerstoffgehaltsregelung des Abgases der Biomasse-Heizanlage der Fig. 1 bis 24 mit einem KI-Modell;

Fig. 31

zeigt ein Verfahren zur wetterprognoseabhängigen Ein-/Aus-Regelung der Biomasse-Heizanlage der Fig. 1 bis 24 mit einem KI-Modell;

Fig. 32

zeigt ein Verfahren zur wetterabhängigen Leistungsregelung der Biomasse-Heizanlage der Fig. 1 bis 24 mit einem KI-Modell.

Fig. 33

zeigt ein Verfahren zur Leistungsregelung der Biomasse-Heizanlage der Fig. 1 bis 24 mit einem KI-Modell;

Fig. 34

zeigt ein Verfahren zur Primärluftregelung der Biomasse-Heizanlage der Fig. 1 bis 24 mit einem KI-Modell.

The biomass heating system according to the invention is explained in more detail below in exemplary embodiments and individual aspects with reference to the figures of the drawing:

1

shows a three-dimensional overview of a biomass heating system according to an embodiment of the invention;

2

shows a cross-sectional view through the biomass heating system 1 , which was taken along a section line SL1 and which is shown viewed from the side view S;

3

also shows a cross-sectional view through the biomass heating system of FIG 1 with an illustration of the flow path, wherein the cross-sectional view was taken along a section line SL1 and is shown viewed from the side S;

4

shows a partial view of the 2 , which has a combustion chamber geometry of the boiler 2 and 3 represents;

figure 5

shows a sectional view through the boiler or the combustion chamber of the boiler along the vertical section line A2 of FIG 4 ;

6

shows a three-dimensional sectional view of the primary combustion zone of the combustion chamber with the rotary grate 4 ;

7

shows according to 6 an exploded view of the combustion chamber bricks;

8

shows a top view of the rotary grate with rotary grate elements seen from the section line A1 of FIG 2 ;

9

shows the rotary grate of 2 in the closed position, with all rotary grate elements aligned horizontally or closed;

10

shows the rotary grate of 9 in the state of a partial cleaning of the rotary grate in ember maintenance mode;

11

shows the rotary grate of 9 in the state of universal cleaning, which is preferably carried out during a plant standstill;

12

Figure 12 shows an exploded oblique view of an exemplary recirculation device with combustor bricks surrounding a primary combustion zone;

13

FIG. 12 shows an exposed semi-transparent oblique view of the recirculation device of FIG 12 ;

14

shows a side view of the recirculation device 5 of FIG figures 12 and 13 ;

15

shows a schematic block diagram that the flow in the respective individual components of the biomass heating system and the recirculation device Figures 12 to 14 indicates;

16

shows an exposed semi-transparent oblique view of a recirculation device of a further embodiment;

17

shows a schematic block diagram that the flow in the respective individual components of a biomass heating system and the recirculation device 16 disclosed according to another embodiment.

18

shows a schematic block diagram of a device and a method with exemplary components of the biomass heating system Figures 1 to 17 , this being based on a local machine learning approach;

Figures 19a, 19b and 19c

show schematic block diagrams of devices and methods with exemplary components of the biomass heating system of FIG Figures 1 to 17 , this being based on a core machine learning approach;

20

Fig. 12 shows a schematic diagram showing a basic concept for a neuron, which is a possible basis of machine learning;

21

shows a schematic diagram representing a three-layer neural network obtained by summarizing in 21 shown neurons is formed;

22

shows a block diagram which represents a data flow or a sequence of a machine learning of a biomass heating system and the application of the learned model to the biomass heating system;

Figures 23a and 23b

show examples of learning curves when learning/training the model

24

Figure 12 shows example weight data including neural network weights.

25

shows a combustion operation or an operating method of the biomass heating system Figures 1 to 24 ;

26

shows a schematic diagram of an application of machine learning for the classification of fuel with various input parameters or sensor data;

Figures 27a, 27b and 27c

show according to sensor data from sensors, which for the application of 26 be used;

28

shows a method for material identification / classification of fuel of the biomass heating system Figures 1 to 24 with an AI model;

29

shows a method for fuel bed height control of the biomass heating system Figures 1 to 24 with an AI model;

30

shows a method for lambda control or for oxygen content control of the exhaust gas of the biomass heating system Figures 1 to 24 with an AI model;

31

shows a method for weather forecast-dependent on / off control of the biomass heating system Figures 1 to 24 with an AI model;

32

shows a method for weather-dependent power control of the biomass heating system Figures 1 to 24 with an AI model.

Figure 33

shows a method for power control of the biomass heating system Figures 1 to 24 with an AI model;

34

shows a method for primary air control of the biomass heating system Figures 1 to 24 with an AI model.

DESCRIPTION OF EXEMPLARY EMBODIMENTS

Various embodiments of the present disclosure are disclosed below with reference to the accompanying drawings, by way of example only. However, embodiments and terms used therein are not intended to limit the present disclosure to particular embodiments, and it should be construed to include various modifications, equivalents, and/or alternatives according to the embodiments of the present disclosure.

If more general terms are used in the description for features or elements shown in the figures, it is intended that not only the specific feature or element is disclosed in the figures for the person skilled in the art, but also the more general technical teaching.

With respect to the description of the figures, the same reference numbers can be used in the individual figures to refer to similar or technically corresponding elements. Furthermore, for the sake of clarity, more elements or features can be shown with reference symbols in individual detailed or sectional views than in the overview views. It can be assumed that these elements or features are also disclosed accordingly in the overview presentations, even if they are not explicitly listed there.

It is to be understood that a singular form of a noun corresponding to an object may include one or more of the things, unless the context in question clearly indicates otherwise.

In the present disclosure, an expression such as "A or B", "at least one of "A or/and B" or "one or more of A or/and B" can include any possible combination of features listed together. Expressions such as "first ", "secondary", "primary" or "secondary" used herein represent and do not limit various elements regardless of their order and/or importance. When an element (e.g., a first element) is described as being "operably" or "communicatively" coupled or connected to another element (e.g., a second element), the element may be directly connected to the other element become or are connected to the other element via another element (e.g. a third element).

For example, a phrase "configured for" (or "configured for") as used in the present disclosure may be replaced with "suitable for," "suitable for," "adapted for," "made for," "capable of," or "designed for." depending on what is technically possible. Alternatively, in a particular situation, a phrase "device configured to" or "set up to" may mean that the device can operate in conjunction with another device or component, or perform a corresponding function.

All sizes specified in "mm" are to be understood as a size range of +- 1 mm around the specified value, unless another tolerance or other ranges are explicitly stated.

It should be noted that the present individual aspects, for example the rotary grate, the combustion chamber or the filter device, are disclosed here separately from or separately from the biomass heating system as individual parts or individual devices. It is therefore clear to the person skilled in the art that individual aspects or parts of the system are also disclosed here on their own. In the present case, the individual aspects or parts of the system are disclosed in particular in the sub-chapters marked by brackets. It is envisaged that these individual aspects can also be claimed separately.

Furthermore, for the sake of clarity, not all features and elements are individually identified in the figures, especially if they are repeated. Rather, the elements and features are each designated as examples. Analogous or identical elements are then to be understood as such.

(biomass heating system)

1 shows a three-dimensional overview of the biomass heating system 1 according to an embodiment of the invention.

The arrow V in the figures indicates the front view of the plant 1, and the arrow S in the figures indicates the side view of the plant 1.

The biomass heating system 1 has a boiler 11 which is mounted on a base 12 of the boiler. The boiler 11 has a boiler housing 13, for example made of sheet steel.

In the front part of the boiler 11 there is a combustion device 2 (not shown), which can be reached via a first maintenance opening with a closure 21 . A rotary mechanism mount 22 for a rotary grate 25 (not shown) supports a rotary mechanism 23 with which drive forces can be transmitted to bearing axles 81 of the rotary grate 25 .

In the central part of the boiler 11 there is a heat exchanger 3 (not shown), which can be reached from above via a second maintenance opening with a closure 31 .

In the rear of the vessel 11 is an optional filter assembly 4 (not shown) having an electrode 44 (not shown) suspended by an insulating electrode support 43 and powered by an electrode supply line 42 . The exhaust gas of the biomass heating system 1 is discharged via an exhaust gas outlet 41 which is arranged downstream of the filter device 4 (fluidically) in terms of flow. A fan can be provided here.

Behind the boiler 11, a recirculation device 5 is provided which a part of the flue or exhaust gas via recirculation channels 51, 53 and 54 and flaps 52 for cooling the combustion process and reuse when Combustion process recirculated. This recirculation device 5 will later with reference to Figures 12 to 17 explained in detail.

Furthermore, the biomass heating system 1 has a fuel supply 6, with which the fuel is conveyed in a controlled manner to the combustion device 2 in the primary combustion zone 26 from the side onto the rotary grate 25. The fuel supply 6 has a cell wheel sluice 61 with a fuel supply opening 65, the cell wheel sluice 61 having a drive motor 66 with control electronics. An axle 62 driven by the drive motor 66 drives a transmission mechanism 63 which can drive a fuel feed screw 67 (not shown) so that the fuel in a fuel feed channel 64 is fed to the combustion device 2 .

In the lower part of the biomass heating system 1 there is an ash removal device 7 which has an ash discharge screw 71 in an ash discharge channel which is operated by a motor 72 .

The biomass heating system 1 also has a control device 100 . This control device 100 is provided with a conventional processor, volatile and non-volatile memory (for example (S)RAM, ROM, flash and/or cache memory), and various interfaces. Analogue or digital inputs and outputs can be provided as interfaces. For example, CAN bus interfaces, 0-10V analog inputs or 4-20 mA analog inputs/outputs for sensors and actuators and/or RS-232 interfaces can be provided. In addition, the control device can preferably (optionally) have at least one interface with an Internet protocol (IP, Ethernet, WLAN) based on the known standards. The control device can thus preferably communicate via the Internet with the data processing devices installed remotely from the biomass heating system 1 .

With the possibility of communication to remote data processing devices or a central server, the control device 100 can represent part of a distributed system for machine learning, which later in relation to the figures 19 is explained in more detail.

Furthermore, the control device 100 can have a keyboard and/or a display for displaying operating data. The display can also have a so-called touch function, with which an operator can make entries on the display.

In addition to the control device 100, a plurality of sensors for detecting physical and/or chemical parameters of the biomass heating system 1 are provided. Examples of such sensors are in relation to the 2 described in more detail.

One of the sensors that may be communicatively coupled to the controller 100 may be a boiler temperature sensor 115 . A combustion chamber 24 or boiler tubes 32 (cf. 2 ) are at least partially covered by a heat exchange medium 38 (cf. 2 ), for example (heating) water. The boiler temperature sensor 115 preferably measures or senses the temperature of the heat exchange medium 38 in the boiler 11 at a location representative of an average temperature of the heat exchange medium 38 in the boiler 11 .

The temperature detected by the boiler temperature sensor 115 is communicated to the control device 100 (preferably as a signal, for example as a voltage signal, as a current signal or as a digital signal), with which the control device 100 can calculate the temperature (which may still have to be calculated from the signal, for example the voltage of 1 volt corresponding to 10 degrees Celsius above zero) is available for further processing.

The control device can store the temperature detected by boiler temperature sensor 115 in a memory (permanent or volatile) and/or use the temperature as training data for machine learning.

The above regarding the boiler temperature sensor 115 and the detected temperature (as a detected physical variable) can also be applied to other sensors and physical or chemical variables, in particular to the sensors which are described later with reference to FIG 2 and 12 to be discribed. as Sensors can be used, in particular, sensors of the fuel bed height or ember bed height 86, the lambda probe 112, the exhaust gas temperature sensor 111, the vacuum sensor 113, the heating water temperature sensor 114.

In addition, the actuators of the biomass heating system 1 can also be communicatively connected to the control device 100 . For example, the air valves 52 of the recirculation device 5, the ignition device 201, the motors 231 and 66, the electrostatic filter 4 or the electrostatic precipitator 4 (e.g. its electrode voltage), the ash removal 7 or its motor 72, the fuel supply 6 with its Cellular wheel sluice 61 or its drive motor 66 or the cleaning device 9 with its drive 91 can be controlled by the control device 100 .

The control device 100 can thus regulate the biomass heating system 1 . At least one detected physical/chemical variable of at least one sensor of the biomass heating system 1 is communicated to the control device 100, the biomass heating system 1 uses this variable(s) to calculate a control response, the control response in turn for setting at least one actuator of the biomass Heating system 1 is used. Due to the setting of the at least one actuator, the physical/chemical processes in the biomass heating system 1 (in particular those of combustion) are in turn influenced, which in turn is detected by the at least one sensor. This closes at least one control loop. Due to the large number of possible control tasks of the control device 100, the control device 100 can also control more than one control circuit of the biomass heating system at the same time.

In contrast to conventional biomass heating systems, the regulation of the present biomass heating system 1 is based on artificial intelligence, which in turn was trained with previously recorded sensor data from at least one biomass heating system 1 or can also be optimized “live” during operation. The present biomass heating system 1 is thus able to achieve better control results and also to solve control tasks which would otherwise regularly require human intervention during operation or which are error-prone without AI or the use of an AI model 104 . Examples of this are given with reference to the 26 ff. described.

2 shows a cross-sectional view through the biomass heating system 1 of FIG 1 , which was taken along a section line SL1 and which is shown viewed from the side S. In the corresponding 3 , which has the same cut as 2 represents, for the sake of clarity, the flows of the flue gas, and fluidic cross-sections are shown schematically. to 3 it should be noted that individual areas compared to the 2 are shown grayed out. This is only for clarity 3 and the visibility of the flow arrows S5, S6 and S7.

From left to right are in 2 the combustion device 2, the heat exchanger 3 and an (optional) filter device 4 of the boiler 11 are provided. The boiler 11 is mounted on the boiler base 12 and has a multi-walled boiler housing 13 in which water or another fluid heat exchange medium 38 can circulate. A water circulation device 14 with a pump, valves, lines, etc. is provided for the supply and removal of the heat exchange medium.

The combustion device 2 has a combustion chamber 24 in which the combustion process of the fuel takes place in the core. The combustion chamber 24 has a multi-part rotary grate 25, which will be explained in more detail later, on which the fuel bed 28 rests. The multi-part rotary grate 25 is rotatably mounted by means of a plurality of bearing axles 81 .

Further referring to 2 and 3 For example, the primary combustion zone 26 of the combustor 24 is encompassed by (a plurality of) combustor brick(s) 29 , whereby the combustor bricks 29 define the geometry of the primary combustion zone 26 . The cross-section of the primary combustion zone 26 (for example) along horizontal section line A1 is substantially oval (for example 380mm +/- 60mm x 320mm +/- 60mm; it should be noted that some of the above size combinations may also result in a circular cross-section). The arrow S1 schematically reproduces the flow from the secondary air nozzle 291, this flow (this is shown purely schematically) having a swirl induced by the secondary air nozzles 291 in order to improve the mixing of the flue gas.

The secondary air nozzles 291 are designed in such a way that they introduce the secondary air (preheated by the combustion chamber bricks 29) tangentially into the combustion chamber 24 with its oval cross section there. This creates a flow S1 with vortices or twists, which runs roughly spirally or helically upwards. In other words, a spiral flow running upwards and rotating about a vertical axis is formed.

The secondary air nozzles 291 are thus aligned in such a way that they introduce the secondary air tangentially into the combustion chamber 24—viewed in the horizontal plane. In other words, the secondary air nozzles 291 are each provided as an inlet for the secondary air that is not aligned with the center of the combustion chamber. Incidentally, such a tangential entry can also be used with a circular combustion chamber geometry.

All the secondary air nozzles 291 are aligned in such a way that they cause either a right-handed or a left-handed flow. In this respect, each secondary air nozzle 291 can contribute to the formation of the eddy currents, with each secondary air nozzle 291 having a similar orientation. In relation to the above, it should be noted that in exceptional cases individual secondary air nozzles 291 can also be arranged in a neutral (centered) or counter-rotating (opposite orientation) manner, although this may degrade the aerodynamic efficiency of the arrangement.

The combustion chamber bricks 29 form the inner lining of the primary combustion zone 26, store heat and are directly exposed to the fire. The combustion chamber stones 29 thus also protect the other material of the combustion chamber 24 , for example cast iron, from the direct effect of the flames in the combustion chamber 24 . The combustion chamber stones 29 are preferably adapted to the shape of the grate 25 . The combustion chamber bricks 29 also have secondary air or recirculation nozzles 291, which recirculate the flue gas into the primary combustion zone 26 for renewed participation in the combustion process and in particular for cooling as required. The secondary air nozzles 291 are not aligned with the center of the primary combustion zone 26, but are aligned acentrically in order to cause a swirl of the flow in the primary combustion zone 26 (ie a swirling and turbulent flow, which will be explained in more detail later). The combustion chamber bricks 29 will be explained in more detail later. Insulation 311 is provided at the boiler tube entrance. The oval cross-sectional shape of the primary combustion zone 26 (and the nozzle) and the length and position of the secondary air nozzles 291 favor the formation and maintenance of a turbulent flow, preferably up to the ceiling of the combustion chamber 24.

A secondary combustion zone 27 adjoins the primary combustion zone 26 of the combustion chamber 26, either at the level of the combustion chamber nozzles 291 (from a functional or combustion-related point of view) or at the level of the combustion chamber nozzle 203 (from a purely structural or constructional point of view) and defines the radiant part of the combustion chamber 26. In the radiant part, the flue gas produced during combustion releases its thermal energy mainly through thermal radiation, in particular to the heat exchange medium, which is located in the two left-hand chambers for the heat exchange medium 38 . The corresponding flue gas flows are in 3 indicated purely by way of example by the arrows S2 and S3. These turbulent flows may also contain slight backflows or other turbulences, which are not represented by the purely schematic arrows S2 and S3. However, the basic principle of the development of the flow in the combustion chamber 24 is clear and can be calculated by a person skilled in the art based on the arrows S2 and S3.

Caused by the injection of secondary air, pronounced twisting or rotating or turbulent flows develop in the isolated or limited combustion chamber 24 . The oval combustion chamber geometry 24 in particular contributes to the fact that the turbulent flow can develop undisturbed or optimally.

After exiting the nozzle 203, which bundles these turbulent flows again, there are candle-flame-shaped rotary flows S2, which can advantageously reach up to the combustion chamber ceiling 204, so that the available space in the combustion chamber 24 is better utilized. The turbulent flows are concentrated in the center of the combustion chamber A2 and make ideal use of the volume of the secondary combustion zone 27 . Furthermore, the constriction, which represents the combustion chamber nozzle 203 for the turbulent flows, reduces the rotational flows, with which turbulences are generated to improve the mixing of the air/flue gas mixture. Cross-mixing therefore takes place through the constriction or constriction through the combustion chamber nozzle 203 . However, the rotational momentum of the flows is at least partially maintained above the combustion chamber nozzle 203, which maintains the propagation of these flows up to the combustion chamber ceiling 204.

The secondary air nozzles 291 are integrated into the elliptical or oval cross-section of the combustion chamber 24 in such a way that, due to their length and their orientation, they induce turbulent flows which cause the flue gas/secondary air mixture to rotate and thereby (again in combination with the combustion chamber nozzle 203 positioned above improved) enable complete combustion with minimal excess air and thus maximum efficiency. This is also in the Figures 19 to 21 illustrated.

The secondary air supply is designed in such a way that it cools the hot combustion chamber bricks 29 by flowing around them and the secondary air itself is preheated in return, whereby the combustion rate of the flue gases is accelerated and the complete combustion even at extreme partial load (e.g. 30% the nominal load) is ensured.

The first maintenance opening 21 is insulated with an insulating material such as Vermiculite ^™ . The present secondary combustion zone 27 is set up in such a way that burnout of the flue gas is ensured. The special geometric design of the secondary combustion zone 27 will be explained in more detail later.

After the secondary combustion zone 27, the flue gas flows into the heat exchange device 3, which has a bundle of boiler tubes 32 provided parallel to one another. The flue gas now flows downwards in the boiler tubes 32, as in 3 indicated by the arrows S4. This part of the flow can also be referred to as the convection part, since the heat dissipation of the flue gas takes place essentially on the boiler tube walls via forced convection. Due to the temperature gradients in the heat exchanger medium, for example in the water, caused in the boiler 11, natural convection of the water occurs, which promotes thorough mixing of the boiler water.

Spring turbulators 36 and spiral or band turbulators 37 are arranged in the boiler tubes 32 in order to improve the efficiency of the heat exchange device 4 . This will be explained in more detail later.

The outlet of the boiler tubes 32 opens into the turning chamber 35 via the turning chamber inlet 34. The turning chamber 35 is sealed off from the combustion chamber 24 in such a way that no flue gas from the turning chamber 35 can flow directly back into the combustion chamber 24. However, a common (removal) transport path for the combustion residues is provided, which can occur in the entire flow area of the boiler 11. If the filter device 4 is not provided, the flue gas is discharged upwards again in the boiler 11 . The other case of the optional filter device 4 is in the 2 and 3 shown. In the process, the flue gas is fed back up into the filter device 4 after the turning chamber 35 (cf. arrows S5), which in the present example is an electrostatic filter device 4. Flow screens can be provided at the inlet 44 of the filter device 4, which even out the inflow of the flue gas into the filter.

Electrostatic dust filters, also known as electrostatic precipitators, are devices for separating particles from gases that are based on the electrostatic principle. These filter devices are used in particular for the electrical cleaning of exhaust gases used. With electrostatic precipitators, dust particles are electrically charged by a corona discharge of a spray electrode and drawn to the oppositely charged electrode (collecting electrode). The corona discharge takes place on a suitable, charged high-voltage electrode (also known as a discharge electrode) inside the electrostatic precipitator. The (spray) electrode is preferably designed with protruding tips and possibly sharp edges, because the density of the field lines and thus also the electric field strength is greatest there and the corona discharge is thus favored. The opposite electrode (precipitation electrode) usually consists of a grounded section of exhaust pipe that is mounted around the electrode. The degree of separation of an electrostatic precipitator depends in particular on the dwell time of the exhaust gases in the filter system and the voltage between the spray and separation electrodes. The rectified high voltage required for this is provided by a high-voltage generating device (not shown). The high-voltage generation system and the holder for the electrode must be protected from dust and dirt in order to avoid unwanted leakage currents and to extend the service life of system 1.

As in 2 shown is a rod-shaped electrode 45 (which is preferably designed like an elongated, plate-shaped steel spring, cf. 15 ) held approximately centrally in an approximately chimney-shaped interior of the filter device 4. The electrode 45 consists at least largely of high-quality spring steel or chromium steel and is held by an electrode holder 43 via a high-voltage insulator, ie an electrode insulation 46 .

The (spray) electrode 45 hangs downwards into the interior of the filter device 4 so that it can vibrate. The electrode 45 can, for example, vibrate back and forth transversely to the longitudinal axis of the electrode 45 .

A cage 48 simultaneously serves as a counter-electrode and as a cleaning mechanism for the filter device 4. The cage 48 is connected to ground or earth potential. The flue gas or exhaust gas flowing in the filter device 4 is filtered by the prevailing potential difference, cf. the arrows S6, as explained above. in the If the filter device 4 is cleaned, the electrode 45 is de-energized. The cage 48 preferably has an octagonal regular cross-sectional profile, as can be seen, for example, in FIG 13 can be taken. The cage 48 can preferably be laser cut during manufacture.

After exiting the heat exchanger 3 (from its exit), the flue gas flows through the turning chamber 34 into the inlet 44 of the filter device 4.

The (optional) filter device 4 is optionally provided fully integrated in the boiler 11, so that the wall surface facing the heat exchanger 3 and flushed through by the heat exchange medium is also used for heat exchange from the direction of the filter device 4, with which the efficiency of the system 1 is further improved. In this way, at least part of the wall of the filter device 4 can be flushed with the heat exchange medium, with the result that at least part of this wall is cooled with boiler water.

The cleaned exhaust gas flows out of the filter device 4 at the filter outlet 47, as indicated by the arrows S7. After leaving the filter, part of the exhaust gas is returned to the primary combustion zone 26 via the recirculation device 5 . This will also be explained in more detail later. This waste gas or flue gas intended for recirculation can also be referred to as “Rezi” or “Rezi gas” for short. The remaining part of the exhaust gas is conducted out of the boiler 11 via the exhaust gas outlet 41 .

An ash discharge 7 is arranged in the lower part of the boiler 11. The ash that has been separated and fallen out, for example from the combustion chamber 24, the boiler tubes 32 and the filter device 4, is discharged laterally out of the boiler 11 via an ash discharge screw 71.

In 2 and 3 further sensors are shown which are at least communicatively connected to the control device 100 . (Physical and/or chemical) variables of the biomass heating system 1 are recorded with the sensors.

An exhaust gas temperature sensor 111 is provided downstream of the exit of the heat exchanger 3 . This measures a temperature of the exhaust gas or flue gas after it has flowed through the heat exchanger 3 .

A conventional temperature sensor or also a PT-100 or PT-1000 sensor can be used as the exhaust gas temperature sensor 111, which is provided in the wall of the exhaust gas duct or protrudes into the exhaust gas duct. With the help of the exhaust gas temperature sensor 111, the temperature of the exhaust gas can be determined in degrees Celsius.

The exhaust gas temperature sensor 111 can be provided before or after the optional filter device 4, for example. Likewise, for example, the exhaust gas sensor 111 can be provided in front of the exhaust gas outlet 41 . Furthermore, more than one exhaust gas temperature sensor 111 can also be provided in order to increase the accuracy of the measurement or also to provide metrological redundancies. For example, an exhaust gas temperature sensor 111 can be provided directly after the outlet of the heat exchanger 3 and a further exhaust gas temperature sensor 111 can be provided after the filter device 4 .

At least one lambda probe 112 is also provided. It is intended as a sensor for the lambda control of the biomass heating system 1 . At least one physical/chemical variable that enables the combustion process in the boiler 11 to be regulated is recorded with the lambda probe. The lambda probe 112 enables an O2 content measurement or an oxygen content measurement of the exhaust gas or the flue gas after the combustion chamber 24.

A lambda probe can usually compare the residual oxygen content in the exhaust gas with the oxygen content of a reference, usually the current atmospheric or ambient air. From this, the combustion air ratio λ (ratio of combustion air to fuel) can be determined and set. It can be two Measuring principles are used: Voltage of a solid electrolyte (Nernst probe) and change in resistance of a ceramic (resistance probe).

In the present use in the biomass heating system, the lambda probe 112 can measure the oxygen content of the exhaust gas (e.g. in vol%) and an optimal mixture can thus be controlled on the boiler 11, preferably by means of an AI model, in order to avoid an oversupply of cooling supply air or carbon monoxide (with unused residual calorific value) arising as a result of a lack of oxygen, which would "steal" energy from the heating system.

For the at least one lambda probe 112 are in 2 two possible installation positions are suggested. One is located adjacent to the inlet 33 of the heat exchanger 3 (cf. 2 , above, center) and the other is located in the exhaust gas outlet 41 and thus after the outlet of the heat exchanger 3 (cf. 2 , top right). In general, the lambda probe 112 can be provided at any position in the exhaust gas duct of the boiler 11, as long as it can measure the exhaust gas or flue gas.

However, the greater the distance selected between the flame in the combustion chamber 24 and the lambda probe 112, the more difficult it becomes to regulate the boiler 11 because of the dead time that then occurs. Therefore, it is preferable to mount the probe as close to the combustion chamber 24 as possible. With the signal from the lambda probe 112, the supply of primary air into the combustion chamber and the amount of fuel supplied can be regulated via the control device 100, for example.

An (optional) vacuum sensor 113 or pressure difference sensor 113 is also provided. This vacuum sensor 113 measures the (negative) pressure in the combustion chamber 24, for example in the unit [mPas], or the differential pressure of the combustion chamber 24 to the ambient air pressure. The primary air (and optionally the secondary air) is sucked into the combustion chamber 24 for combustion via the negative pressure.

An (optional) return (or flow) temperature sensor 114 or a heating water temperature sensor 114 is also provided. This is provided, for example, in the return or in the flow of a conventional water circulation device 14 and detects the temperature of the heating water in the water circuit in which the boiler 11 is provided. The heat exchange medium 38 is preferably the heating water.

In this way, the temperature of the heat exchange medium 38 in or outside the boiler can be detected with the previously explained boiler temperature sensor 115 or with the heating water temperature sensor 114 (preferably a return temperature sensor 114).

A fuel bed height sensor 116 (shown here in the figures without example mechanics) detects the height of the fuel bed 28 above the grate and thus an amount of the fuel, for example the wood chips, on the grate 25. An example of such a sensor is a mechanical version in the EP 3 789 670 B1 in relation to theirs 17 and 18 described, referred to above. Alternatively, the fuel bed height sensor 116 can be provided as an ultrasonic sensor, for example.

A combustion chamber temperature sensor 117 is also provided. This detects a temperature of the combustion chamber 24, for example in degrees Celsius. The combustion chamber temperature sensor 117 can be provided at the outlet of the combustion chamber 24 or also in the combustion chamber 24 .

It should be noted that the locations of the sensors of the 2 and 3 may deviate from the locations shown, as deemed appropriate by those skilled in the art. For example, the combustion chamber temperature can also be recorded at a different location.

The combustor 24 and boiler 11 of this embodiment were calculated using CFD simulations. Furthermore, practical experiments were carried out to confirm the CFD simulations. The starting point for the considerations Calculations for a 100 kW boiler, but considering a power range from 20 to 500 kW.

A CFD simulation (CFD = Computational Fluid Dynamics = numerical fluid mechanics) is the spatially and temporally resolved simulation of flow and heat conduction processes. The flow processes can be laminar and/or turbulent, accompanied by chemical reactions, or it can be a multi-phase system. CFD simulations are therefore well suited as a design and optimization tool. In the present invention, CFD simulations were used to optimize the fluidic parameters in such a way that the objects of the invention listed above are achieved. In particular, as a result, the mechanical design and dimensioning of the boiler 11, the combustion chamber 24, the secondary air nozzles 291 and the combustion chamber nozzle 203 were largely defined by the CFD simulation and also by associated practical experiments. The simulation results are based on a flow simulation taking heat transfer into account.

The components of the biomass heating system 1 and the boiler 11 listed above, which are results of the CFD simulations, are described in more detail below.

(combustion chamber)

The design of the combustor shape is important in order to be able to meet the task requirements. The shape and geometry of the combustion chamber are intended to ensure the best possible turbulent mixing and homogenization of the flow across the cross section of the flue gas duct, minimization of the combustion volume, and a reduction in excess air and the recirculation ratio (efficiency, operating costs), a reduction in CO and CxHx Emissions, NOx emissions, dust emissions, a reduction in local temperature peaks (fouling and slagging) and a reduction in local flue gas velocity peaks (material stress and erosion) can be reached.

the 4 showing a partial view of the 2 is, and the figure 5 , which is a sectional view through the boiler 11 along the vertical section line A2, represent a combustion chamber geometry that meets the above-mentioned requirements for biomass heating systems over a wide power range of, for example, 20 to 500 kW. Incidentally, the vertical section line A2 can also be understood as the middle or central axis of the oval combustion chamber 24 . In the 4 those sensors are drawn, such as these in relation to the 2 and 3 were explained.

$$\mathrm{BK2}=300\mathrm{mm}+-50\mathrm{mm},\mathrm{vorzugsweise}+-30\mathrm{mm};$$

$$\mathrm{BK3}=430\mathrm{mm}+-80\mathrm{mm},\mathrm{vorzugsweise}+-40\mathrm{mm};$$

$$\mathrm{BK4}=538\mathrm{mm}+-80\mathrm{mm},\mathrm{vorzugsweise}+-50\mathrm{mm};$$

$$\mathrm{BK5}=\left(\mathrm{BK3}-\mathrm{BK2}\right)/2=\mathrm{bspw}.65\mathrm{mm}+-30\mathrm{mm},\mathrm{vorzugsweise}+-20\mathrm{mm};$$

$$\mathrm{BK6}=307\mathrm{mm}+-50\mathrm{mm},\mathrm{vorzugsweise}+-20\mathrm{mm};$$

$$\mathrm{BK7}=82\mathrm{mm}+-20\mathrm{mm},\mathrm{vorzugsweise}+-20\mathrm{mm};$$

$$\mathrm{BK8}=379\mathrm{mm}+-40\mathrm{mm},\mathrm{vorzugsweise}+-20\mathrm{mm};$$

$$\mathrm{BK9}=470\mathrm{mm}+-50\mathrm{mm},\mathrm{vorzugsweise}+-20\mathrm{mm};$$

$$\mathrm{BK10}=232\mathrm{mm}+-40\mathrm{mm},\mathrm{vorzugsweise}+-20\mathrm{mm};$$

$$\mathrm{BK11}=380\mathrm{mm}+-60\mathrm{mm},\mathrm{vorzugsweise}+-30\mathrm{mm};$$

$$\mathrm{BK12}=460\mathrm{mm}+-80\mathrm{mm},\mathrm{vorzugsweise}+-30\mathrm{mm}.$$

The in the Figures 3 and 4 The dimensions given and determined via CFD calculations and practical experiments for an exemplary boiler with approx. 100 kW are as follows: $$\mathrm{BK1}=172\mathrm{mm}+-40\mathrm{mm},\mathrm{preferably}+-17\mathrm{mm};$$

$$\mathrm{BK2}=300\mathrm{mm}+-50\mathrm{mm},\mathrm{preferably}+-30\mathrm{mm};$$

$$\mathrm{BK3}=430\mathrm{mm}+-80\mathrm{mm},\mathrm{preferably}+-40\mathrm{mm};$$

$$\mathrm{BK4}=538\mathrm{mm}+-80\mathrm{mm},\mathrm{preferably}+-50\mathrm{mm};$$

$$\mathrm{BK5}=\left(\mathrm{BK3}-\mathrm{BK2}\right)/2=\mathrm{e.g}.65\mathrm{mm}+-30\mathrm{mm},\mathrm{preferably}+-20\mathrm{mm};$$

$$\mathrm{BK6}=307\mathrm{mm}+-50\mathrm{mm},\mathrm{preferably}+-20\mathrm{mm};$$

$$\mathrm{BK7}=82\mathrm{mm}+-20\mathrm{mm},\mathrm{preferably}+-20\mathrm{mm};$$

$$\mathrm{BK8}=379\mathrm{mm}+-40\mathrm{mm},\mathrm{preferably}+-20\mathrm{mm};$$

$$\mathrm{BK9}=470\mathrm{mm}+-50\mathrm{mm},\mathrm{preferably}+-20\mathrm{mm};$$

$$\mathrm{BK10}=232\mathrm{mm}+-40\mathrm{mm},\mathrm{preferably}+-20\mathrm{mm};$$

$$\mathrm{BK11}=380\mathrm{mm}+-60\mathrm{mm},\mathrm{preferably}+-30\mathrm{mm};$$

$$\mathrm{BK12}=460\mathrm{mm}+-80\mathrm{mm},\mathrm{preferably}+-30\mathrm{mm}.$$

All dimensions and sizes are to be understood as examples only.

With these values, both the geometries of the primary combustion zone 26 and the secondary combustion zone 27 of the Combustion chamber 24 optimized. The specified size ranges are ranges with which the requirements are (approximately) fulfilled as well as with the specified exact values.

A chamber geometry of the primary combustion zone 26 and the combustion chamber 24 (or an inner volume of the primary combustion zone 26 of the combustion chamber 24) can preferably be defined using the following basic parameters:
A volume with an oval horizontal base measuring 380 mm +- 60 mm (preferably +-30 mm) x 320 mm +- 60 mm (preferably +-30 mm), and a height of 538 mm +- 80 mm ( preferably +- 50 mm).

The size information given above can also be applied to boilers in other output classes (e.g. 50 kW or 200 kW) scaled in relation to one another.

As a further development, the volume defined above can have an upper opening in the form of a combustion chamber nozzle 203, which is provided in the secondary combustion zone 27 of the combustion chamber 24, which has a combustion chamber slope 202 protruding into the secondary combustion zone 27, which preferably contains the heat exchange medium 38. Combustion chamber slope 202 reduces the cross section of secondary combustion zone 27. Combustion chamber slope 202 is inclined by an angle k of at least 5%, preferably by an angle k of at least 15% and even more preferably by at least an angle k of 19% with respect to an imaginary Horizontal or straight combustion chamber ceiling H (cf. the dashed horizontal line H in 4 ) intended.

In addition, a combustion chamber cover 204 is provided, likewise inclined in the direction of the inlet 33 . The combustion chamber 24 in the secondary combustion zone 27 thus has the combustion chamber ceiling 204 which is provided inclined upwards in the direction of the inlet 33 of the heat exchanger 3 . This combustion chamber ceiling 204 extends in section 2 at least mostly straight or rectilinear and inclined. The angle of inclination of the straight or flat combustion chamber ceiling 204 can preferably be 4 to 15 degrees relative to the (fictitious) horizontal.

With the combustion chamber ceiling 204, a further (ceiling) slope is provided in the combustion chamber 24 in front of the inlet 33, which forms a funnel together with the combustion chamber slope 202. This funnel turns the swirling or eddy flow directed upwards to the side and deflects this flow more or less horizontally. Due to the already turbulent upward flow and the funnel shape in front of the inlet 33, it is ensured that all heat exchanger tubes 32 or boiler tubes 32 are flown evenly, whereby an evenly distributed flow of the flue gas in all boiler tubes 32 is ensured. This optimizes the heat transfer in the heat exchanger 3 considerably.

In particular, the combination of the vertical and horizontal inclines 203, 204 in the secondary combustion zone in combination as the inflow geometry in the convective boiler can achieve a uniform distribution of the flue gas over the convective boiler tubes.

The combustion chamber slope 202 serves to homogenize the flow S3 in the direction of the heat exchanger 3 and thus the flow through the boiler tubes 32. This causes the flue gas to be distributed as evenly as possible to the individual boiler tubes in order to optimize the heat transfer there.

In detail, the combination of the inclines with the inflow cross section of the boiler rotates the flue gas flow in such a way that the flue gas flow or the flow rate is distributed as evenly as possible over the respective boiler tubes 32 .

In the prior art, there are often combustors with a rectangular or polygonal combustor and nozzle, but the irregular shape of the combustor and the nozzle and their interaction is another obstacle to a uniform air distribution and a good mixture of air and fuel and thus good burnout, as has been recognized herein. In particular, with an angular geometry of the combustion chamber, flow threads or preferential flows arise, which disadvantageously lead to an uneven flow of the heat exchanger tubes 32 .

Therefore, in the present case, the combustion chamber 24 is provided without dead corners or dead edges.

It was thus recognized in the present case that the geometry of the combustion chamber (and the entire course of flow in the boiler) plays a decisive role in the considerations for optimizing the biomass heating system 1 . For this reason, the oval or round basic geometry described here without dead corners was chosen (in contrast to the usual rectangular or polygonal or purely cylindrical shapes). In addition, this basic geometry of the combustion chamber and its structure with the dimensions / dimension ranges given above have been optimized for a 100 kW boiler. These dimensions/dimensional ranges are selected in such a way that, in particular, different fuels (wood chips and pellets) with different qualities (for example with different water contents) can be burned with a very high level of efficiency. This was the result of practical tests and CFD simulations.

In particular, the primary combustion zone 26 of the combustion chamber 24 can comprise a volume which preferably has an oval or approximately circular horizontal cross-section on the outer circumference (such a cross-section is shown in 2 marked with A1 as an example). This horizontal cross section can also preferably represent the base area of the primary combustion zone 26 of the combustion chamber 24 . The combustion chamber 24 can have an approximately constant cross section over the height indicated by the double arrow BK4. In this respect, the primary combustion zone 24 can have an approximately oval-cylindrical volume. Preferably, the side walls and base (grate) of the primary combustion zone 26 may be perpendicular to one another. The bevels 203 , 204 described above can be provided as integrated walls of the combustion chamber 24 be, wherein the slopes 203, 204 form a funnel, which opens in the inlet 33 of the heat exchanger 33 and has the smallest cross-section there.

The term "approximately" is used above because of course individual notches, design-related deviations or small asymmetries can be present, for example at the transitions of the individual combustion chamber bricks 29 to one another. However, these minor deviations only play a subordinate role in terms of flow technology.

The horizontal cross section of the combustion chamber 24 and in particular of the primary combustion zone 26 of the combustion chamber 24 can also preferably be regular. Further, the horizontal cross-section of the combustor 24, and particularly the primary combustion zone 26 of the combustor 24, may preferably be a regular (and/or symmetrical) ellipse.

In addition, the horizontal cross section (the outer circumference) of the primary combustion zone 26 can be made constant over a predetermined height (for example, 20 cm).

An oval-cylindrical primary combustion zone 26 of the combustion chamber 24 is thus provided in the present case, which, according to CFD calculations, enables a significantly more uniform and better air distribution in the combustion chamber 24 than in the case of rectangular combustion chambers of the prior art. The lack of dead spaces also avoids zones in the combustion chamber with poor air flow, which increases efficiency and reduces slag formation.

Likewise, the nozzle 203 in the combustion chamber 24 is designed as an oval or approximately circular constriction in order to further optimize the flow conditions. The swirl of the flow in the primary combustion zone 26 explained above, which is caused by the specially designed secondary air nozzles 291 according to the invention, leads to a roughly helical or spiral upward flow pattern, with an equally oval or approximately circular nozzle Favors the course of the flow and does not interfere with the usual rectangular nozzles. This optimized nozzle 203 bundles the flue gas-air mixture flowing upwards rotating and ensures better mixing, preservation of the eddy currents in the secondary combustion zone 27 and thus complete combustion. This also minimizes the excess air required. This improves the combustion process and increases efficiency.

The combination of the secondary air nozzles 291 explained above and the eddy currents induced thereby with the optimized nozzle 203 serves in particular to bundle the flue gas/air mixture rotating upwards. This ensures at least approximately complete combustion in the secondary combustion zone 27.

Thus, a turbulent or swirling flow is bundled through the nozzle 203 and directed upwards, with the result that this flow extends further upwards than is usual in the prior art. As can be seen by the person skilled in the art from the laws of physics relating to the angular momentum, this is due to the reduction in the distance of the swirling air flow to the rotation or swirl center axis, which is forced by the nozzle 203 (compare analogously to the physics of the pirouette effect).

In addition, the course of the flow in the secondary combustion zone 27 and from the secondary combustion zone 27 to the boiler tubes 32 is presently optimized, as explained in more detail below.

The combustion chamber slope 202 of 4 , which without a reference number in the 2 and 3 can be seen and where the combustion chamber 25 (or its cross-section) tapers at least approximately linearly from bottom to top, according to CFD calculations ensures that the flue gas flow in the direction of the heat exchange device 4 is made more uniform, which means that its efficiency can be improved. The horizontal cross-sectional area of the combustion chamber 25 tapers from the beginning to the end of the combustion chamber slope 202, preferably by at least 5%. The combustion chamber slope 202 is on the side of the combustion chamber 25 to the Heat exchange device 4 is provided, and is provided rounded at the point of maximum taper. Parallel or straight combustion chamber walls without a taper (so as not to impede the flue gas flow) are common in the prior art. In addition, there is, individually or in combination, the combustion chamber cover 204, which extends obliquely upwards towards the inlet 33 to the horizontal and diverts the turbulent flows in the secondary combustion zone 27 laterally, thereby equalizing their flow velocity distribution.

The inflow or deflection of the flue gas flow in front of the tube bundle heat exchanger is designed in such a way that an uneven flow of the tubes is avoided as far as possible, whereby temperature peaks in individual boiler tubes 32 can be kept low and thus the heat transfer in the heat exchanger 4 can be improved (best possible use of the heat exchanger surfaces). . As a result, the efficiency of the heat exchange device 4 is improved.

In detail, the gaseous volume flow of the flue gas is conducted through the inclined combustion chamber wall 203 at a uniform speed (even in the case of different combustion states) to the heat exchanger tubes or the boiler tubes 32. This effect is further intensified by the sloping combustion chamber ceiling 204, with a funnel effect being brought about. The result is a uniform heat distribution of the heat exchanger surfaces affecting the individual boiler tubes 32 and thus an improved use of the heat exchanger surfaces. The exhaust gas temperature is thus reduced and the efficiency increased. The flow distribution is particularly at the in the 3 shown indicator line WT1 much more evenly than in the prior art. The line WT1 represents an entry surface for the heat exchanger 3. The indicator line WT3 indicates an exemplary cross-sectional line through the filter device 4, in which the flow is set up as homogeneously as possible or is approximately evenly distributed over the cross-section of the boiler tubes 32 (due to of flow screens at the entrance of the filter device 4 and due to the geometry of the turning chamber 35). A uniform flow through the filter device 3 and the last boiler pass minimizes strand formation and thereby also optimizes the separation efficiency of the filter device 4 and the heat transfer in the biomass heating system 1.

Furthermore, an ignition device 201 is provided in the lower part of the combustion chamber 25 on the fuel bed 28 . This can cause initial ignition or re-ignition of the fuel. The ignition device 201 can be a glow igniter. The ignition device is advantageously stationary and offset horizontally to the side relative to the location at which the fuel is introduced.

Furthermore, the lambda probe 112 can (optionally) be provided after the exit of the flue gas (i.e., after S7) from the filter device. A control device 100 can use the lambda probe to identify the respective calorific value. The lambda probe 112 can thus enable the ideal mixing ratio between the fuels and the oxygen supply to be regulated. Despite different fuel qualities, the result is high efficiency and higher efficiency.

A good and flexible control of the (operating) parameters of the combustion and of the biomass heating system 1 serves to maintain or improve the processes in the biomass heating system 1 explained above. The adjustment of the actuators based on the sensory feedback by means of AI or trained AI model significantly improves the processes explained above.

This in figure 5 The fuel bed 28 shown shows a rough fuel distribution due to the feeding of the fuel from the right side of the figure 5 . This fuel bed 28 is flown from below with a mixture of flue gas and fresh air, which is provided by the recirculation device 5 . This flue gas/fresh air mixture is advantageously pre-tempered and has the ideal quantity (mass flow) and the ideal mixing ratio, such as the control device 100 regulates on the basis of various measured values recorded by sensors and associated air valves 52 . The burning behavior of the fuel bed can be influenced in particular with the height of the fuel bed, which is detected by the fuel bed height sensor 116 . According to experiments, the height of the fuel bed has been shown to be a good indicator that can be used for the AI model, which amount of fuel is available for a burn. In order to optimize the combustion process, the height of the fuel bed is also a relevant variable to be controlled for the biomass heating system 1.

Next is in the 4 and 5 a combustion chamber nozzle 203 is shown in which a secondary combustion zone 27 is provided and which accelerates and focuses the flue gas flow. As a result, the flue gas flow is better mixed and can burn more efficiently in the post-combustion zone 27 or secondary combustion zone 27 . The area ratio of the combustion chamber nozzle 203 is in a range from 25% to 45%, but is preferably 30% to 40%, and is, for example for a 100 kW biomass heating system 1, ideally 36% +/- 1% (ratio of the measured input area to the measured exit area of the nozzle 203).

Consequently, the above information on the combustion chamber geometry of the primary combustion zone 26 together with the geometry of the secondary air nozzles 291 and the nozzle 203 represent an advantageous development of the present disclosure.

(combustion chamber bricks)

the 6 shows a three-dimensional sectional view (obliquely from above) of the primary combustion zone 26 and the isolated part of the secondary combustion zone 27 of the combustion chamber 24 with the rotary grate 25, and in particular the special design of the combustion chamber bricks 29. The 7 shows according to 6 an exploded view of the combustion chamber bricks 29. The views of 6 and 7 can preferably with the dimensions listed above 4 and 5 be executed. However, this is not necessarily the case.

The chamber wall of the primary combustion zone 26 of the combustor 24 is provided with a plurality of combustor bricks 29 in a modular construction which, among other things, facilitates manufacture and maintenance. The maintenance will facilitated in particular by the possibility of removing individual combustion chamber bricks 29.

Form-fitting grooves 261 and projections 262 (in 6 to avoid redundancies in the figures, only a few of these are designated by way of example) are provided in order to create a mechanical and largely airtight connection, in order in turn to prevent the ingress of disturbing external air. Preferably, every two at least largely symmetrical combustion chamber bricks (with the possible exception of the openings for the secondary air or the recirculated flue gas) form a complete ring. Furthermore, three rings are preferably stacked on top of one another in order to form the primary combustion zone 26 of the combustion chamber 24 which is oval-cylindrical or alternatively at least approximately circular (the latter is not shown).

Three further combustion chamber bricks 29 are provided as the upper closure, with the annular nozzle 203 being supported by two retaining bricks 264 which are placed on the upper ring 263 in a form-fitting manner. All bearing surfaces 260 have grooves 261 either for mating projections 262 and/or for the insertion of suitable sealing material.

The mounting stones 264, which are preferably symmetrical, can preferably have an inwardly inclined bevel 265 in order to make it easier for fly ash to be swept away onto the rotary grate 25.

The lower ring 263 of the combustion chamber bricks 29 rests on a base plate 251 of the rotary grate 25 . Ash is increasingly deposited on the inner edge between this lower ring 263 of the combustion chamber bricks 29 , which advantageously independently and advantageously seals this transition during operation of the biomass heating system 1 .

The (optional) openings for the recirculation nozzles 291 or secondary air nozzles 291 are provided in the middle ring of the combustion chamber bricks 29 . There they are Secondary air nozzles 291 are provided at least approximately at the same (horizontal) height of the combustion chamber 24 in the combustion chamber bricks 29 .

Three rings of combustor bricks 29 are provided here, as this represents the most efficient way of manufacture and also of maintenance. Alternatively, 2, 4 or 5 such rings can also be provided.

The combustion chamber bricks 29 are preferably made of high-temperature silicon carbide, which makes them very wear-resistant.

The combustion chamber bricks 29 are provided as shaped bricks. The combustion chamber bricks 29 are shaped in such a way that the interior volume of the primary combustion zone 26 of the combustion chamber 24 has an oval horizontal cross section, which means that dead corners or dead spaces, which are usually not optimally flown through by the flue gas/air mixture, are avoided by an ergonomic shape, whereby the fuel present there is not is optimally burned. Due to the present shape of the combustion chamber bricks 29, the flow of primary air through the grate 25, which also suits the distribution of the fuel over the grate 25, and the possibility of unhindered turbulent flows is improved; and consequently the efficiency of combustion is improved.

The oval horizontal cross section of the primary combustion zone 26 of the combustion chamber 24 is preferably a point-symmetrical and/or regular oval with the smallest inside diameter BK3 and the largest inside diameter BK11. These measurements were the result of optimizing the primary combustion zone 26 of the combustor 24 using CFD simulation and field testing.

(rotary grate)

8 shows a plan view of the rotary grate 25 from above seen from the section line A1 of FIG 2 .

The supervision of 8 can preferably be designed with the dimensions listed above. However, this is not necessarily the case.

The rotary grate 25 has the base plate 251 as a base element. A transition element 255 is provided in a roughly oval-shaped opening in the base plate 251, which bridges a gap between a first rotary grate element 252, a second rotary grate element 253 and a third rotary grate element 254, which are rotatably mounted. Thus, the rotary grate 25 is provided as a rotary grate with three individual elements, i. i.e. this can also be referred to as a triple rotary grate. Air holes are provided in the rotary grate elements 252, 253 and 254 for primary air to flow through.

The rotary grate elements 252, 253 and 254 are flat and heat-resistant metal plates, for example made of cast metal, which have an at least largely planar configured surface on the upper side and are connected to the bearing axles 81 on the lower side, for example via intermediate mounting elements. When viewed from above, rotating grate elements 252, 253 and 254 have curved and complementary sides or contours.

In particular, the rotating grate elements 252, 253, 254 can have mutually complementary and curved sides, with the second rotating grate element 253 preferably having concave sides to the adjacent first and third rotating grate elements 252, 254, and preferably the first and third rotating grate elements 252, 254 to the respective second rotary grate element 253 has a convex side. This improves the crushing function of the rotary grate elements, since the length of the fracture is increased and the forces acting to break (similar to scissors) act in a more targeted manner.

$$\mathrm{DR2}=350\mathrm{mm}+-60\mathrm{mm},\mathrm{bevorzugt}+-20\mathrm{mm}$$

The rotary grate elements 252, 253 and 254 (and their enclosure in the form of the transition element 255) have an approximately oval outer shape when viewed together, which in turn avoids dead corners or dead spaces in which suboptimal combustion could take place or ash could accumulate could accumulate undesirably. The optimal dimensions of this outer shape of the rotating grate elements 252, 253 and 254 are in 8 denoted by the double arrows DR1 and DR2. Preferably, but not exclusively, DR1 and DR2 are defined as follows: $$\mathrm{DR1}=288\mathrm{mm}+-40\mathrm{mm},\mathrm{preferred}+-20\mathrm{mm}$$

$$\mathrm{DR2}=350\mathrm{mm}+-60\mathrm{mm},\mathrm{preferred}+-20\mathrm{mm}$$

These values have turned out to be optimal values (ranges) in the CFD simulations and the following practical test. These dimensions correspond to those of 4 and 5 . These dimensions are particularly advantageous for the combustion of different fuels or the fuel types wood chips and pellets (hybrid firing) in a power range of 20 to 200 kW.

The rotary grate 25 has an oval combustion surface, which is more favorable for the fuel distribution, the air flow through the fuel and the combustion of the fuel than a conventional rectangular combustion surface. The combustion surface 258 is formed at the core by the surfaces of the rotary grate members 252, 253 and 254 (in the horizontal state). The combustion surface is therefore the upward-pointing surface of the rotary grate elements 252, 253 and 254. This oval combustion surface advantageously corresponds to the fuel support surface if this is applied or pushed laterally onto the rotary grate 25 (cf. arrow E of 9 , 10 and 11 ). In particular, the fuel can be supplied from a direction which is parallel to a longer central axis (main axis) of the oval combustion surface of the rotary grate 25 .

The first rotary grate element 252 and the third rotary grate element 254 can preferably be configured identically in their combustion surface 258 . Furthermore, the first rotary grate element 252 and the third rotary grate element 254 can be identical or structurally identical to one another. For example, this is in 9 1, with the first rotating grate element 252 and the third rotating grate element 254 having the same shape.

Furthermore, the second rotary grate element 253 is arranged between the first rotary grate element 252 and the third rotary grate element 254 .

The rotary grate 25 is preferably provided with an approximately point-symmetrical, oval combustion surface 258 .

Likewise, the rotary grate 25 can form an approximately elliptical combustion surface 258, with DR2 being the dimensions of its major axis and DR1 being the dimensions of its minor axis.

Furthermore, the rotary grate 25 can have an approximately oval combustion surface 258 which is axisymmetric with respect to a central axis of the combustion surface 258 .

Furthermore, the rotary grate 25 can have an approximately circular combustion surface 258, which entails minor disadvantages in terms of fuel supply and distribution.

Two motors or drives 231 of the rotary mechanism 23 are also provided, with which the rotary grate elements 252, 253 and 254 can be rotated accordingly. More about the special function and the advantages of the present rotary grate 25 is later with reference to the figures 9 , 10 and 11 described.

In particular in the case of pellet and wood chip heating systems (and in particular in the case of hybrid biomass heating systems), failures due to slag formation in the combustion chamber 24, in particular on the rotary grate 25, can increasingly occur. Slag is always produced during a combustion process when temperatures above the ash melting point are reached in the embers. The ash then softens, sticks together, and cools to form a solid, and often dark-colored, slag. This process, also referred to as sintering, is undesirable in the biomass heating system 1 since the accumulation of slag in the combustion chamber 24 leads to a malfunction can: it switches itself off. The combustion chamber 24 usually has to be opened and the slag removed.

The ash melting range (this extends from the sintering point to the flow point) depends very significantly on the fuel used. Spruce wood, for example, has a critical temperature of around 1,200 °C. However, the ash melting range of a fuel can also be subject to strong fluctuations. Depending on the quantity and composition of the minerals contained in the wood, the behavior of the ash changes during the combustion process.

Another factor that can influence slag formation is the transport and storage of the wood pellets or chips. This is because they should reach the combustion chamber 24 as undamaged as possible. If the wood pellets have already crumbled when they enter the combustion process, this increases the density of the ember bed. The result is more slag formation. In particular, the transport from the storage room to the combustion chamber 24 is of importance here. Particularly long distances, as well as curves and angles, lead to damage or abrasion of the wood pellets.

Another factor relates to the control of the combustion process. So far, efforts have been made to keep the temperatures rather high in order to achieve the best possible burnout and low emissions. An optimized combustion chamber geometry and geometry of the combustion zone 258 of the rotary grate 25 makes it possible to keep the combustion temperature at the grate lower and high in the area of the secondary air nozzles 291, and thus to reduce slag formation on the grate.

In addition, the resulting slag (and also the ash) can advantageously be removed due to the special shape and the functionality of the present rotary grate 25 . This will now be related to the figures 9 , 10 and 11 explained in more detail.

the figures 9 , 10 and 11 show a three-dimensional view of the rotary grate 25 with the base plate 251, the first rotary grate element 252, the second rotary grate element 253 and the third rotary grate element 254. The views of FIG 9 , 10 and 11 be able preferably correspond to the dimensions listed above. However, this is not necessarily the case.

This view shows the rotary grate 25 as a free slide-in part with rotary grate mechanism 23 and drive(s) 231. The rotary grate 25 is mechanically provided in such a way that it can be individually prefabricated in the manner of the modular system, and as a slide-in part inserted into a provided elongated opening of the boiler 11 and can be installed. This also facilitates the maintenance of this wear-prone part. The rotary grate 25 can thus preferably have a modular design, in which case it can be quickly and efficiently removed and reinserted as a complete part with rotary grate mechanism 23 and drive 231 . The modularized rotary grate 25 can thus also be assembled and disassembled using quick-release fasteners. In contrast, prior art rotary grates are typically permanently mounted and thus difficult to maintain or assemble.

The drive 231 can have two separately controllable electric motors. These are preferably provided on the side of the rotating grate mechanism 23 . The electric motors can have reduction gears. Furthermore, end stop switches can be provided which provide end stops for the end positions of the rotating grate elements 252, 253 and 254 respectively.

The individual components of the rotary grate mechanism 23 are intended to be exchangeable. For example, the gears are provided to be plugged. This facilitates maintenance and also a side change of the mechanics during assembly, if necessary.

In the rotary grate elements 252, 253 and 254 of the rotary grate 25, the openings 256 already mentioned are provided. The rotary grate elements 252, 253 and 254 can each be rotated by at least 90 degrees, preferably at least 120 degrees, even more preferably by 170 degrees, via their respective bearing axles 81, which are driven by the drive 231, in this case the two motors 231, via the rotary mechanism 23 Degrees are rotated about the respective bearing or axis of rotation 81. The maximum angle of rotation can be 180 degrees, or a little less than 180 degrees, like the one Allow rust lips 257. The rotary mechanism 23 is set up in such a way that the third rotary grate element 254 can be rotated individually and independently of the first rotary grate element 252 and the second rotary grate element 243, and that the first rotary grate element 252 and the second rotary grate element 243 are rotated together and independently of the third rotary grate element 254 be able. The rotary mechanism 23 can be provided accordingly, for example by means of running wheels, toothed or drive belts and/or gears.

The rotary grate elements 252, 253 and 254 can preferably be produced as a cast grate with a laser cut in order to ensure precise shape retention. This in particular in order to define the air flow through the fuel bed 28 as precisely as possible and to avoid disturbing air currents, for example strands of air at the edges of the rotating grate elements 252, 253 and 254.

The openings 256 in the rotating grate elements 252, 253 and 254 are arranged in such a way that they are small enough for the usual pellet material and/or the usual wood chips not to fall through and large enough for the fuel to flow well with air can be. In addition, the openings 256 are dimensioned large enough that they can be blocked by ash particles or impurities (e.g. no stones in the fuel).

9 12 now shows the rotary grate 25 in the closed position, with all rotary grate elements 252, 253 and 254 being horizontally aligned or closed. This is the position in normal operation. The uniform arrangement of the large number of openings 256 ensures a uniform flow through the fuel bed 28 (this is shown in 9 not shown) on the rotary grate 25 ensured. In this respect, the optimal combustion state can be produced here. The fuel is applied to the rotary grate 25 from the direction of arrow E; to that extent the fuel from the right side of the 9 pushed up onto the rotary grate 25.

During operation, ash and/or slag accumulates on the rotary grate 25 and in particular on the rotary grate elements 252, 253 and 254. The rotary grate 25 can be cleaned efficiently with the present rotary grate 25 .

10 shows the rotary grate in the state of a partial cleaning of the rotary grate 25 in ember maintenance mode. To this end, only the third rotating grate element 254 is rotated. Because only one of the three rotary grate elements is rotated, the embers are retained on the first and second rotary grate elements 252, 253, while at the same time the ash and slag can fall out of the combustion chamber 24 downwards. As a result, no external ignition is required to resume operation (this saves up to 90% ignition energy). A further consequence is a reduction in wear on the ignition device (for example an ignition rod) and a saving in electricity. Ash cleaning can also advantageously take place during operation of the biomass heating system 1 .

10 also shows a state of ember maintenance during a (often already sufficient) partial cleaning. The operation of the system 1 can thus advantageously take place more continuously, which means that, in contrast to the usual full cleaning of a conventional grate, there is no need for lengthy, complete ignition, which can take several tens of minutes.

In addition, any potential slag formation or slag accumulation on the two outer edges of the third rotary grate element 254 is broken up as it rotates, with the curved outer edges of the third rotary grate element 254 not only shearing off over a greater overall length than with conventional rectangular elements of the stand of the technique, but also with an uneven distribution of movement in relation to the outer edge (there is more movement in the middle than at the bottom and top edges). The breaker function of the rotating grate 25 is thus significantly strengthened.

In 10 grate lips 257 (on both sides) of the second rotary grate element 253 can be seen. This grate lips 257 are set up such that the first rotary grate element 252 and the third rotary grate element 254 in the closed state of this on the Rest on top of the grate lips 257, and thus the rotary grate elements 252, 253 and 254 are provided to each other without a gap and are thus provided sealing. This avoids strands of air and undesired uneven primary air flows through the bed of embers. This advantageously improves the efficiency of the combustion.

11 shows the rotary grate 25 in the state of universal cleaning, which is preferably carried out during a plant standstill. All three rotary grate elements 252, 253 and 254 are rotated, with the first and second rotary grate elements 252, 253 preferably being rotated in the opposite direction to the third rotary grate element 254. This enables the rotary grate 25 to be completely emptied on the one hand and the ash and slag now broken up on four odd outside edges. In other words, an advantageous 4-fold breaker function is realized. The above in relation to 9 What is explained with regard to the geometry of the outer edges also applies with regard to 10 .

In summary, the present rotary grate 25 realizes in addition to normal operation (cf. 9 ) advantageously two different types of cleaning (cf. 10 and 11 ), whereby the partial cleaning allows a cleaning during the operation of the plant 1.

In comparison to this, commercially available rotary grate systems are not ergonomic and, due to their rectangular geometry, have disadvantageous dead spots in which the primary air cannot optimally flow through the fuel, with the result that air strands can form. At these corners, there is also an accumulation of slag formation. These points ensure poorer combustion with poorer efficiency.

The existing simple mechanical structure of the rotary grate 25 makes it robust, reliable and durable.

(recirculation device)

In order to optimize the recirculation device 5 briefly mentioned above, CFD simulations, further considerations and practical tests were again carried out. The flue gas recirculation described below was provided for a biomass heating system.

In the calculations, for example, a 100 kW boiler was simulated in the nominal load operating case with a load range of 20 to 500 kW with different fuels (e.g. wood chips with 30% water content). In the present case, slight soiling or soot (so-called fouling with a thickness of 1 mm) was also taken into account for all surfaces in contact with flue gas. The emissivity of such a fouling layer was set at 0.6.

The result of this optimization and the accompanying considerations is in the Figures 12 to 17 shown. the Figures 12 to 14 show different views of the recirculation device 5, which in the Figures 1 to 3 is evident.

12 12 shows an exposed oblique view of the recirculation device 5 with the combustion chamber bricks 29 surrounding the primary combustion zone 26. FIG. 13 shows an exposed semi-transparent oblique view of the recirculation device 5 of FIG 12 . the 14 shows a side view of the recirculation device 5 of FIG figures 12 and 13 . The arrow S of Figures 12 to 14 corresponds to the arrow S of the 1 , which indicates the direction of the side view of the biomass heating system 1.

The recirculation device 5 is described below with reference to figures 12 , 13 , 14 and 15 described in more detail.

The recirculation device 5 has a recirculation inlet 53 with a recirculation inlet channel 531 and a recirculation inlet channel divider 532 . The recirculation inlet 53 and the recirculation inlet duct 531 are downstream of a fan 15 (cf. 3 ) at the flue gas outlet of the biomass heating system 1 the heat exchanger 3 or after the (optional) filter device 4 is provided. The recirculation inlet duct splitter 532 can branch the flue gas to be recirculated or the rezi gas into a primary recirculation duct 56 and an optional secondary recirculation duct 57 . If there is no secondary recirculation, no recirculation inlet channel splitter 532 is required either.

The primary recirculation duct 56 opens out via an air valve 52, in this case a rotary slide valve 52, for example, in a primary mixing chamber 542. In the primary mixing chamber 542 opens out via a further air valve 52, in this case a rotary slide valve 52, for example, also a primary air duct 58, which in turn has a primary air inlet 581 for space, for example - or fresh air, correspondingly referred to as primary fresh air. The primary air duct 58 can have a primary air sensor 582 (for example for detecting the temperature and/or the oxygen content of the primary fresh air and/or the flow rate of the air).

Unmixed primary air, i. H. Fresh air or ambient air, in the primary mixing chamber 542, in which the ambient air is mixed according to the valve position of the air valves 52 with the recirculated flue gas from the primary recirculation channel 56. A primary mixing duct 54 is provided downstream of the primary mixing chamber 542, in which the mixture of primary (fresh) air and flue gas is further mixed. The primary mixing chamber 542 with its valves 52 and the primary mixing channel 54 together form a primary mixing unit 5a.

The secondary recirculation channel 57 opens into a secondary mixing chamber 552 via an air valve 52, in this case a rotary slide valve 52, for example. A secondary air channel 59, which in turn has a secondary air inlet 591 for secondary fresh air, also opens into the secondary mixing chamber 552 via a further air valve 52, in this case a rotary slide valve 52, for example . The secondary air channel 59 can have a secondary air sensor 592 (for example for detecting the temperature and/or the oxygen content of the secondary air and/or the flow rate of the air).

The fresh secondary air, i. H. Ambient air in the secondary mixing chamber 552, in which the ambient air is mixed according to the valve position of the air valves 52 with the recirculated flue gas from the secondary recirculation channel 57. A secondary mixing duct 55 is provided downstream of the secondary mixing chamber 552, in which the mixture of secondary fresh air and flue gas is further mixed. The secondary mixing chamber 552 with its valves 52 and the secondary mixing channel 55 form the secondary mixing unit 5b.

The position of the four air valves 52 is set in each case by means of a valve setting actuator 521, which can be an electric motor, for example. In the 12 only one of the four valve position actuators 521 is designated for reasons of clarity. These actuators can be controlled by the control device 100 to regulate the respective air volume and are communicatively connected to the control device 100 .

The primary mixing channel 54 has a minimum length L1. The minimum length L1 is, for example, at least 700 mm from the beginning of the primary mixing duct 54 at the exit from the primary mixing chamber 542 to the end of the primary mixing duct 54. It has been shown that the length L1 of the primary mixing duct 54 can be longer for good mixing, preferably at least 800 mm mm, ideally 1200 mm. In addition, the length L1 should preferably not exceed 2000 mm, for example, for design and printing reasons. The primary mixing duct 54 can have an inlet funnel at its upstream beginning, which tapers towards the end of the primary mixing duct 54 . The flow at the upstream beginning of the channel 54 is thus bundled in the middle and mixes even better, since stranding can occur, particularly on the upper side of the channel 54, due to thermal differences. This formation of strands is advantageously counteracted by the narrowing of the primary mixing channel 54 at its beginning.

The (optional) secondary mixing channel 55 has a minimum length L2. The minimum length L2 is, for example, at least 500 mm from the beginning of the secondary mixing duct 55 at the outlet from the secondary mixing chamber 552 to the end of the secondary mixing duct 55. It has been shown that the length L2 of the secondary mixing duct 55 can be longer for good mixing, preferably at least 600 mm mm, ideally 1200 mm. The length L2 should also not exceed 2000 mm, for example, for design and printing reasons. The secondary mixing channel 55 can also have an inlet funnel at its upstream start, which tapers in the direction of the downstream end of the secondary mixing channel 55 .

The primary mixing duct 54 and the (optional) secondary mixing duct 55 can be designed with a rectangular cross section with a respective inner width of 160 mm +- 30 mm (vertical) / 120 mm +- 30 mm (vertical) and an inner thickness (horizontal) of 50 mm +- be executed 15 mm. Due to this design of the primary mixing duct 54 and the secondary mixing duct 55 each as a long, flat duct lying against the heat exchanger 3 and the combustion device, several advantageous effects are achieved. On the one hand, the mixture of flue gas and primary (fresh) air/secondary (fresh) air is advantageously preheated before it is burned. For example, a mixture that has a temperature of +25 degrees Celsius downstream of the primary mixing chamber 542 can have a temperature that is 15 degrees Celsius higher at the downstream end of the primary mixing channel 54 in the case of nominal load. On the other hand, the cross section and the longitudinal extent are selected to be large enough that the mixing also continues after the mixing chambers 542, 552, which causes an improvement in the homogenization of the flow. In doing so, the flow is provided with sufficient path for a further mixing of the flow, which is already turbulent at the beginning of the path.

In other words, the elongate primary mixing duct 54 provides a path for further mixing after the primary mixing chamber 542, with the primary mixing chamber 542 being purposefully designed to create significant turbulence at the beginning of the path. The optional inlet funnel of channels 54, 55 can also contribute to this.

The two lengths L1 and L2 can preferably match within a certain tolerance (+−10 mm).

The recirculated flue gas, which was previously well mixed with "fresh" primary air, is fed from below to the rotary grate 25 via a primary passage 541 . This mixture of recirculated flue gas and fresh primary air (i.e., the primary air for the combustor 24) enters the primary combustion zone 26 of the combustor 24 through its openings 256 . In this respect, the primary recirculation for recirculating the mixture of flue gas and primary fresh air is provided in such a way that it enters the primary combustion zone 26 from below.

Via an (optional) secondary passage 551 and a subsequent ring channel 50 (cf. 13 ) around the combustion chamber bricks 29, the recirculated flue gas, which has previously been well mixed with "fresh" secondary air, i.e. secondary fresh air (or if secondary recirculation is omitted, with primary (fresh) air), is directed to the (also optional) recirculation or secondary air nozzles 291 fed. As explained, the secondary air nozzles 291 are not aligned with the center of the primary combustion zone 26, but rather are aligned off-centre in order to cause a twist in the flow running upwards out of the primary combustion zone 26 into the secondary combustion zone 27 (i.e. an upward Vortex flow with a vertical spin axis). In this respect, the secondary recirculation for recirculating the mixture of flue gas and secondary fresh air can be provided at least partially in the secondary combustion zone 27 .

the figures 13 and 14 show according to the 12 the course of the flows of the air, the recirculated flue gas and the flue gas-air mixtures in the recirculation device 5 using the (schematic) flow arrows S8 to S16. The arrows S1 to S16 indicate the fluidic configuration, ie the course of the flow of the various gases or moving masses in the biomass heating system 1. Many of the present components or features are fluidically connected, this being indirectly (ie via other components ) or can be immediate.

Like in the 13 and the 14 As can be seen in each case, the flue gas, which flows out of the heat exchanger 3 and the optional filter device 4 after the heat exchange, enters the recirculation inlet channel 531 of the recirculation device 5 through the recirculation inlet 5 (cf. arrow S8). After an (optional) division of the flue gas flow by an (optional) recirculation inlet duct splitter 532, the flue gas from the primary recirculation flows through the primary recirculation duct 56 (see arrow S10), depending on the position of one of the adjustable air valves 52, into the primary mixing chamber 541, in which the flue gas with of the primary fresh air is mixed, which also flows through the primary air duct 58, depending on the position of another of the adjustable air valves 52, into the primary mixing chamber 541 (see arrow S12).

As a result, a mixed flow (cf. arrow S14) of flue gas and fresh primary air is created in the primary mixing duct 54, in which these two components mix advantageously due to the turbulence and the length of the primary mixing duct 54. At the end of the primary mixing duct 54, a homogeneous mixture of flue gas and primary fresh air has formed, which flows through the primary passage 541 to the primary combustion zone 26 (cf. arrow S16).

If there is a secondary recirculation (similar to the primary recirculation in terms of flow), the flue gas, after being divided in the recirculation inlet duct divider 532, flows through the secondary recirculation duct 57 via a further adjustable air valve 52 into the secondary mixing chamber 552 (see arrow S9), in which the flue gas with the over the secondary air duct 59 and a further adjustable valve 52 is also mixed with the secondary fresh air flowing into the secondary mixing chamber 552 (cf. arrow S11). This mixing of the flue gas and the secondary fresh air continues in the secondary mixing duct (see arrow S13), which improves the mixing of both components. The resulting advantageously homogeneous mixture flows through the secondary passage 551 into the annular channel 50 around the combustion chamber bricks 29 and through the recirculation nozzles 291 into the combustion chamber 24 (cf. arrow S15).

The physical/chemical parameters of this recirculation can be recorded by the corresponding sensors 582, 592, etc. Likewise, the recirculation (for example the mixing ratio, flow rates, etc.) can be regulated by means of the corresponding actuators 52, 15, etc.

The schematic block diagram of the 15 shows the above with reference to FIG Figures 12 to 14 Explained flow in the respective individual components of the recirculation device 5, and the biomass heating system 1. In the block diagram 15 Both the primary recirculation and the optional secondary recirculation are shown as a complete circuit. Accordingly, the sensors and actuators of the biomass heating system 1 are located, with which the biomass heating system 1 can be controlled. The recirculation device 5 can also only have a primary recirculation, cf. 17 .

By means of the recirculation of the flue gas, this is basically mixed with fresh air after combustion, the oxygen content in particular being increased, and fed to renewed combustion. This means that combustible residues in the flue gas, which would otherwise be discharged unused through the chimney, can now make a contribution to combustion.

The respective valves 52 with the primary mixing chamber 541 and the (preferably approximately horizontally extending) primary mixing channel 54 form the primary mixing unit 5a. The respective valves 52 with the secondary mixing chamber 552 and the secondary mixing channel 55 can form the secondary mixing unit 5b. Concerning the in 14 hidden parts of the flow guide is on the 3 and the associated explanations.

Next is in 15 the so-called false air entry is also taken into account, which was considered as a disruptive factor in this case. In doing so, leakage air from the environment reaches the combustion chamber 24 via leaks and in particular also the fuel supply, this representing an additional source of air for the combustion that occurs when the mixing ratio of the mixture or mixtures is set is to be taken into account. For this reason, the biomass heating system 1 is preferably set up in the present case in such a way that the secondary air entry in the nominal load operating case is reduced to less than 6%, preferably less than 4%, of the air quantity of the mixture of primary fresh air and recirculated flue gas (and if there is secondary recirculation of the air quantity of the mixture of secondary fresh air and recirculated flue gas and the mixture of primary fresh air and recirculated flue gas).

Incidentally, false air could also disadvantageously get back into the combustion chamber 24 from the further flow path of the flue gas after the combustion, for example via the usual ash removal.

The in the figures 14 and 15 The sensors 111, 112, 113 and 117 shown, as well as 582 and 592, correspond to those above in relation to FIG Figures 1 to 3 and 12 , as well as in relation to 18 described sensors.

(flue gas recirculation of another embodiment)

16 shows an exposed semi-transparent oblique view of a recirculation device of a further embodiment.

In this further embodiment, there is no recirculation as in the embodiment of the secondary air supply 13 provided, but a simple controlled or regulated fresh air supply. This further embodiment is easier and cheaper to produce, but can still have many of the above-mentioned advantages of the embodiment 13 Offer. In particular, the set efficiency goals could also be achieved with this embodiment, as practical tests have shown.

Matching reference numbers of 16 reveal essentially the same teaching of 13 , which is why, in order to avoid repetition, only the differences between the two embodiments will be discussed.

The rotary slide valves of the embodiment of 13 are in the further embodiment 13 been replaced by sliding gate valves. Next takes place in the further embodiment of 16 no secondary mixing of rezi and fresh air takes place, but only the supply (amount) of fresh air to the recirculation nozzles 291 is controlled or regulated. The secondary mixing duct 55 was retained as the secondary temperature control duct 55a, fulfilling the function of temperature control of the fresh air. The secondary temperature control duct 55a is provided along the wall of the boiler 11, whereby the fresh air, which is supplied from the secondary air duct 59, is preheated by the heat of the boiler 11 before the secondary air is introduced into the combustion chamber 24 (see arrow S13a). Correspondingly, the secondary temperature control channel 55a is provided with a rectangular cross section, which has a greater (vertical) height than (horizontal) thickness, whereby the secondary temperature control channel 55a "snuggles up" to the boiler wall and the area for heat exchange is kept large. Preheated secondary air increases combustion efficiency. With regard to the design of the secondary temperature control channel 55a in detail, reference is also made to the statements relating to the secondary mixing channel 55 .

The arrow S15 shows the secondary air flow flowing through the secondary passage 551 into the annular duct 50 around the combustion chamber bricks 29 and through the recirculation nozzles 291 into the combustion chamber 24. This means that not only is the secondary air advantageously further heated, but the combustion chamber bricks 29 are also advantageously cooled, which, for example, slag formation on the combustion chamber bricks is reduced (cf. the above statements on the minimum temperature during slag formation).

The arrows S8 and S10 only indicate the flow of the flue gas downstream of the heat exchanger 3 (or the optional filter device 4) to the primary mixing unit 5a, which in this embodiment is constructed more simply and cheaply.

17 shows a schematic block diagram that the flow in the respective individual components of a biomass heating system and the recirculation device 16 disclosed according to the further embodiment.

Same reference numbers of 17 reveal essentially the same teaching of 15 , which is why, in order to avoid repetition, only the differences are dealt with in the core.

There is no secondary air mixture of fresh air and Rezi gas. In this respect, no secondary mixing chamber 552 and no valve 52 are provided for the Rezi gas. Likewise, the recirculation inlet channel divider 532 is omitted. The secondary mixing channel 55 can be mechanically identical to the embodiment of FIG 15 be, but is functionally not a duct section for mixing fresh air and Rezi gas, but only serves (this is still the same as the embodiment of the 15 ) the pre-temperature control of the fresh air before it is introduced into the combustion chamber 24.

In the further embodiment, the secondary air supply can also be completely dispensed with, in which case the biomass heating system 1 can only be provided with primary recirculation.

The in the 17 drawn sensors 111, 112, 113, and 117, as well as 582 and 592 according to the above in relation to Figures 1 to 3 and 12 , as well as in relation to 18 described.

(Biomass heating system with an AI model or with a control device optimized using machine learning)

The biomass heating system 1 described above is provided with a control device 100 which has artificial intelligence (AI) or an AI model 104 . This AI is prepared by machine learning and allows optimized control of the biomass heating system 1.

Using the methods explained below 18 a local approach (i.e. an on-site approach) of machine learning is explained, while later in relation to the Figures 19a to 19c a centralized approach (ie learning in a central server and execution and data collection on-site) is explained.

18 shows a schematic block diagram with the components of a control of the biomass heating system Figures 1 to 17 .

The regulation includes the three steps of detecting the relevant physical and/or chemical variables in the biomass heating system 1 by sensors, processing the detected variables in the control device 100, and setting the respective actuators of the biomass heating system 1 as a result of the processing

With these three basic steps, a wide variety of applications can be implemented with the support of machine learning. These specific applications include the classification of the fuel (pellets, wood chips, elephant grass, nut shells, etc.), determining how full the combustion chamber 24 is with fuel, calculating the lambda value, optimizing the control of the electrostatic filter unit, controlling the Boiler activity (on/off), the regulation of the power output of the boiler 11 or the control of the air valves of the biomass heating system 1. More information can be found in Figures 26 to 33 .

It should be noted that the chemical and/or physical variables detected by the sensors can also be referred to as raw data or sensor data from a data technology point of view. This sensor data can be analog or digital.

The first step in detecting the relevant physical/chemical parameters in the biomass heating system 1 can be carried out with the aid of the following sensors, cf figures 1 and 17 and the associated description are carried out: exhaust gas temperature sensor 111, lambda probe 112, vacuum sensor or pressure difference sensor 113, return temperature sensor or heating water temperature sensor 114, boiler temperature sensor 115, fuel bed height sensor 116, 86 and/or combustion chamber temperature sensor 117. At least one primary air sensor 582 and/or a secondary air sensor 592 can also be used. Depending on the application, one or more of these sensors 111-117, 582, 592 can be used. This list is also not necessarily complete, for example external sensors or other sources can provide information about relevant measurement variables or input data for machine learning. An example would be a flow sensor for the heating water, a heat quantity sensor for the heat quantity of the heating water circulating in the building, the (current or requested) power requirement of the building, a sensor for determining the current position of the sun or information from a weather forecast (which, for example, is queried via the Internet can become). For example, machine learning could be used to inductively determine the future energy requirements of the building from a weather forecast and carry out a corresponding power control. An air volume sensor (not shown) can also be provided at the exhaust gas outlet 41 or at the outlet or inlet of the blower 15 . A flow sensor for measuring, for example, the flow rate of the boiler water or for measuring the flow of the flue gas can also be considered as a sensor.

Physical quantities are detected by the sensors 111-117 and other sensors, with the result of this detection being transmitted to the control device 100 via a connection 198 . Such a transmission can take place, for example, by means of various signals (e.g. digitally via the CAN bus, or analogously via a 4-20 mA signal). In this respect, the sensors are communicatively connected to the control device 100 . As a result of the detection, the control device 100 is provided with sensor data by the sensors 111-117.

In terms of data technology, this sensor data can either already be obtained in the sensor (and then, for example, be transmitted as digital values to the control device) and/or also only be obtained in control device 100 from the transmitted signals. In the first case, a so-called smart sensor can be used, for example, which transmits a data via a digital interface transmits a specific temperature value or a pressure value as the sensor data. In the second case, for example, a 0-10V voltage signal from the sensor can be converted into digital values by an A/D converter in control device 100, and the digital values can then be converted (for example using a conversion formula) into the sensor data which, for example, indicates a temperature or a represent pressure. For the purposes of the present teaching, it is irrelevant where the (final) sensor data is generated, whether in sensor 111-117 or first in control device 100.

The control device 100 has at least one input interface, at least one output interface, a memory and a processing unit for processing the recorded data in the second step.

A machine learning unit 101 is also provided, which can be provided, for example, as a software module or a separate hardware module in order to carry out machine learning. The function of the machine learning unit 101 will be discussed later in relation to the 22 explained in more detail. The machine learning unit 101 can be provided in the control device 100, as in FIG 18 shown, or also separately from the control device 100, for example in a central processing unit 190 (cf. Figures 19a until c) or as a separate module of control device 100.

In principle, a number of recorded physical quantities, which are stored in the form of sensor data, are selected and then used to train a corresponding model, which in turn is then iteratively evaluated and optimized until the model's prediction accuracy is considered to be sufficient. A sufficiently optimized model is then used as an AI model for the ongoing operation of the biomass heating system.

The control device 100 can also have an optional user interface 102 or a user interface 102 (e.g. a display, a touchscreen, a keyboard, a network device such as a tablet PC or a cell phone, etc.), which is particularly suitable for training with the machine learning unit 101 can be used can. For example, the model can be optimized and evaluated with or without user support.

The control device 100 can include a memory for storing the detected physical and/or chemical quantities and/or the result of the machine learning.

With the result of the processing, the actuators are adjusted in a third step in order to optimize the regulation of the biomass heating system. In this respect, the result of the machine learning is used to influence or directly control the control behavior of the biomass heating system 1 . The actuators include, for example and not exclusively, the valves 52, 52s (or the air flaps), the rotary valve 61, the ignition device 201, the motors of the rotary grate 231, the fuel supply or the drive motor 66, the ash discharge 71, 72, the cleaning drive 91, the filter device 4, the water circulation device 14 and the blower 15. Other actuators not mentioned here can also be controlled. These actuators can be communicatively connected to the control device 100 via a connection 198, for example analog or digital.

The first, second and third step can preferably be carried out iteratively in the sense of a control loop.

Although it is shown that the connection 198 is unidirectional, it can of course also be bidirectional, for example by means of a bus system.

In short, the control device 100, which contains the machine learning unit 101, communicates with various sensors and actuators of the biomass heating system 1. The control device can first carry out machine learning at the processor in order to set an AI. As a result, the AI can then enable optimized control of the biomass heating system 1 .

• Erfassen der zumindest einen relevanten physikalischen/chemischen Größe der Biomasse-Heizanlage 1 durch zumindest einen Sensor 111-117, 582, 592 (oder durch eine Mehrzahl von Sensoren) der Biomasse-Heizanlage 1 zur Bereitstellung von entsprechenden Sensordaten, wobei der Sensor/die Sensoren mit der Steuereinrichtung 100 kommunikativ verbunden ist/sind;
• Verarbeiten der erfassten Größe oder der Größen (bzw. der Sensordaten);
• Einstellen des zumindest eines Aktors (oder der Aktoren) der Biomasse-Heizanlage 1, welcher (welche) mit der Steuereinrichtung (100) kommunikativ verbunden ist (sind) und von dieser angesteuert werden kann (können).

In summary, the regulation is in 18 shown with the following steps:

• Detection of the at least one relevant physical/chemical variable of the biomass heating system 1 by at least one sensor 111-117, 582, 592 (or by a plurality of sensors) of the biomass heating system 1 to provide corresponding sensor data, the sensor/sensors is/are communicatively connected to the controller 100;
• processing the detected size or sizes (or sensor data);
• Adjusting the at least one actuator (or actuators) of the biomass heating system 1, which (which) is (are) communicatively connected to the control device (100) and can be controlled by it.

where is in 18 the (physical and/or chemical) reaction of the system to the setting of the actuators is shown with the dashed arrow "biomass heating system 1". For example, if the setting of air valves 52 changes, the combustion behavior in the combustion chamber will also change, for example the combustion chamber temperature or the lambda value will change.

The step of processing the detected variables includes in control mode, ie during normal operation of the biomass heating system 1, using an AI model (104) with which the setting of various actuators using selected sensor data or at least one filtered variable is calculated. More information can be found in the Figures 20 to 22 .

The AI model is parameterized using machine learning. The AI model is thus trained in a learning mode in which stored quantities or sensor data are used.

This machine learning in the context of the present biomass heating system 1 (in the learning mode) is described in more detail below.

The machine learning has the function of extracting, e.g., a useful rule, a knowledge representation, and/or a determination criterion based on an analysis of a data set input to the device, outputting the determination results, and learning the knowledge (machine learning). A variety of learning techniques are available, which are roughly classified into, for example, "supervised learning", "unsupervised learning" and "reinforcement learning" as discussed above. To implement these techniques, another technique called "deep learning" is available, which involves self-learning to extract feature magnitudes. Although these types of machine learning can use a general-purpose computer or processor, using e.g. GPGPU (General-Purpose Computing on Graphics Processing Units) or computers with high computing power allows for higher processing speed.

First, in supervised learning, a (preferably large) number of data sets of specific inputs and results (labels) are introduced into or supplied to the control device 100, which learns the relationships or features contained in these data sets and inductively creates a model for predicting or estimating the Result learned from the input, ie their relationship. Supervised learning is applicable to this embodiment, for example, for use in classifying the fuel used (e.g., wood chips or pellets). Supervised learning can be realized by using an algorithm such as a neural network (explained in detail later). In this context, the categories of data into which the data sets of the sensor data are to be classified are referred to as “labels”. Thus, labels (in German "labeling" or "category") are the output or the result on which the model is trained. For example, a classification model returns a label as a result. A label is the basis for optimizing the model to a "statement". In many cases, labels are predefined for a specific application.

In unsupervised learning, only input data in a large amount is introduced into a learning unit of the controller 100, which learns a distribution or characteristic of the input data and in turn carries out learning with a device that compresses, classifies, and the introduced data, for example forms without corresponding interactions with a user being required for learning. This makes it possible, for example, to combine features or characteristics seen in these data sets into similar features or characteristics. The result obtained can be used to define certain criteria and to allocate outputs in an optimized way according to the criteria and thus to predict an output. It should be noted that unsupervised learning does not require this additional information, since the associated algorithms operate purely on the data content. Unsupervised machine learning works on data similarities, but not on previously defined categories.

A solution to the problem lying between unsupervised learning and supervised learning, which is referred to as semi-supervised learning, is also applicable to the present biomass heating system 1 . This is true when, for example, only some data serve as records of inputs and outputs and the rest of the data consist of only inputs. In this aspect, learning can be effectively performed by applying data (e.g., image data or simulation data) even without actual combustion in the biomass heating system 1 .

The schematic structure of the 18 FIG. 1 shows a device and a method for local learning (ie on-site learning in, on or near the biomass heating system 1). Consequently, all the components are present on or in the biomass heating system 1 so that the machine learning can be carried out. Supervised learning can be carried out, for example, with the help of a user who accompanies a machine learning learning process on a touch screen.

The biomass heating system 1 can iteratively run through a number of "learning cycles", with the system 1 being in operation and the sensors detecting the required variables and transmitting them to the control device 100 . These captured Variables are stored in the control device 100 and then enable the machine learning described above by means of the learning unit 101. Depending on the type of machine learning, this can be done by means of a user interface 102, for example a display on the biomass heating system 1 or a mobile phone or a tablet PC, which are connected to the control device 100 via a network, also take place with the involvement of the user or not.

Figures 19a, 19b and 19c show schematic block diagrams of devices and methods with exemplary components of the biomass heating system of FIG Figures 1 to 17 , this being in contrast to 18 based on a core machine learning approach. A plurality of biomass heating systems 1 can be connected to a central computing device 190, for example a server 190, via a network connection 199, for example the Internet. The sensors and the actuators correspond to those of the Figures 1 to 17 and the 18 .

The aim of this central configuration is to collect data or variables from a plurality of biomass heating systems 1 and then make them available for learning so that the resulting AI models are based on a broader or better database.

This differed in the central approach or the central configuration of the Figures 19a,b and c from the local approach or the local configuration of the 18 essentially in architecture.

the Figures 19a and 19b show for reasons of clarity only a single biomass heating system 1 and a single control device 100. As in synopsis with Figure 19c can be seen, however, a plurality of biomass heating systems 1 and a plurality of control devices 100 is provided in the central approach, which correspondingly in the Figures 19a and 19b need to be supplemented mentally.

In the Figure 19a and the Figure 19b the biomass heating system 1 with its sensors and actuators (left block) is connected to the control device 100 (middle block) via a connection 198 . The control device 100 is in turn connected to a central computing device 190 or a server 190 (right block) via a network connection 199 (for example an Internet Protocol (IP) connection via the Internet). Corresponding interfaces are, as in 18 , present.

The left block of each Figure 19a and the Figure 19b (the biomass heating system 1) roughly corresponds to the left and right blocks of the 18 . For simplification are in Figure 19a and in Figure 19b the individual details of the left and right blocks 1, 1 der 18 not stated repeatedly. In this respect, on the 18 referenced and their content in the Figure 19a and the Figure 19b incorporated by reference. The left block 1 of Figure 19a and the Figure 19b thus includes the regarding 18 sensors and actuators mentioned as examples. Correspondingly, at least one connection 198 to the control unit 100 (middle block 100 of the Figure 19a ) provided, with which signals from the sensors / actuators of the biomass heating system 1 can be transmitted to the control unit 100 and back.

The controller 100 of the biomass heating system (s) 1 of Figures 19a and 19b has an AI model 104 which is used to control the biomass heating system 1 . Furthermore, in the control device 100, sensor data or detected chemical and/or physical parameters of the biomass heating system 1 are buffered in a memory 105a. These collected chemical and/or physical variables are then transmitted to the central computing device 190 or the server 190, which consequently collects physical/chemical variables from a plurality of control devices 100 and stores them in a memory 105b.

Consequently, with the central approach, Figures 19a,b and c Sensor data or chemical and / or physical parameters of a plurality of biomass heating systems 1 aggregated. A learning unit 101 of the server 190 can thus carry out machine learning based on aggregated chemical and/or physical quantities of a plurality of biomass heating systems.

• Erfassen von zumindest einer chemischen und/oder physikalischen Größe der Biomassen-Heizanlage (1) durch zumindest einen Sensor 111-117, 582, 592 (oder durch eine Mehrzahl von Sensoren) 111-117, 582, 592 der Biomasse-Heizanlage 1, welcher (welche) mit der Steuereinrichtung 100 kommunikativ verbunden ist (sind), wobei Sensordaten zur Verfügung gestellt werden;
• Ansteuern einer Mehrzahl von Aktoren 4, 5, 52, 6, 61, 66, 7, 72, 91, 201, 231 der Biomasse-Heizanlage (1), welche mit der Steuereinrichtung (100) kommunikativ verbunden sind und von dieser angesteuert werden können;
• Abspeichern der von dem zumindest einem Sensor 86, 111-117, 582, 592 erfassten Größe (oder der Mehrzahl von Sensoren erfassten Größen) als Sensordaten in dem Speicher 105a der Steuereinrichtung 100;
• Übertragen der in dem Speicher 105a der Steuereinrichtung 100 gespeicherten Sensordaten über eine Netzwerkverbindung 199 an die zentrale Recheneinheit 190; und
• Übertragen, von der zentralen Recheneinheit 190 an die Steuereinrichtung 100 der Biomasse-Heizanlage 1 über das Netzwerk 199, des KI-Modells 104 zur Regelung der Biomasse-Heizanlage 1, wobei das KI-Modell 104 mittels der maschinellen Lerneinheit 101 der zentralen Recheneinheit 190 durch maschinelles Lernen zumindest unter Verwendung der übertragenen Sensordaten parametriert wird.

From the point of view of a biomass heating system 1, the central approach thus includes the following process steps:
Controlling the biomass heating system 1 using an AI model 104 of the control device 100, which is parameterized by machine learning, wherein controlling the biomass heating system 1 includes the following:

• Detection of at least one chemical and/or physical variable of the biomass heating system (1) by at least one sensor 111-117, 582, 592 (or by a plurality of sensors) 111-117, 582, 592 of the biomass heating system 1, which (which) is (are) communicatively connected to the control device 100, with sensor data being made available;
• Controlling a plurality of actuators 4, 5, 52, 6, 61, 66, 7, 72, 91, 201, 231 of the biomass heating system (1), which are communicatively connected to the control device (100) and can be controlled by it ;
• Storing the quantity detected by the at least one sensor 86, 111-117, 582, 592 (or the quantity detected by the plurality of sensors) as sensor data in the memory 105a of the control device 100;
• Transmission of the sensor data stored in the memory 105a of the control device 100 via a network connection 199 to the central processing unit 190; and
• Transferred from the central processing unit 190 to the control device 100 of the biomass heating system 1 via the network 199, the AI model 104 for controlling the biomass heating system 1, the AI model 104 using the machine learning unit 101 of the central processing unit 190 machine learning is parameterized at least using the transmitted sensor data.

The central processing unit 190 can, for example, query the sensor data or stored variable(s) from the control device 100 at regular intervals, or the control device 1 of the biomass heating system can independently request the variables or sensor data at regular intervals or when other triggering conditions occur transmitted to the central processing unit 190. The transmission can take place using standard network protocols and technologies.

• Erfassen von zumindest einer chemischen und/oder physikalischen Größe der Biomassen-Heizanlage (1) durch zumindest einen Sensor 111-117, 582, 592 (oder eine jeweilige Mehrzahl von Sensoren) der Biomasse-Heizanlagen 1, welcher (welche) mit der Steuereinrichtung 100 kommunikativ verbunden ist (sind), wodurch Sensordaten zur Verfügung gestellt werden;
• Abspeichern der erfassten Größe(n) bzw. Sensordaten in den Speichern 105a der Steuereinrichtungen 100 der Biomasse-Heizanlagen 1 als Sensordaten bzw. Sensordatensatz;
• Übertragen zumindest eines Teils der in den Speichern 105a der Steuereinrichtungen 100 der Biomasse-Heizanlagen 1 gespeicherten Sensordaten über zumindest eine Netzwerkverbindung 199 an die zentrale Recheneinheit 190; und

From a system perspective, the central approach of a distributed system with a plurality of biomass heating systems 1 and with at least one central computing device 190 (or a central server 190) can have the following method steps:

• Detection of at least one chemical and/or physical variable of the biomass heating system (1) by at least one sensor 111-117, 582, 592 (or a respective plurality of sensors) of the biomass heating system 1, which (which) is (are) connected to the control device 100 is (are) communicatively connected, whereby sensor data is made available;
• Saving the detected variable(s) or sensor data in the memories 105a of the control devices 100 of the biomass heating systems 1 as sensor data or sensor data set;
• Transmission of at least part of the sensor data stored in the memories 105a of the control devices 100 of the biomass heating systems 1 via at least one network connection 199 to the central processing unit 190; and

• (Weiteres) Aggregieren der von der Mehrzahl der Biomasse-Heizanlagen 1 übertragenen Sensordaten in der zentralen Recheneinheit, womit die jeweiligen Sensordaten der Biomasse-Heizanlagen 1 zusammengefasst werden;
• Maschinelles Lernen mittels der maschinellen Lerneinheit 101 unter Verwendung der aggregierten Sensordaten, woraus ein parametriertes KI-Modell 104 resultiert;
  • Übertragen des parametrierten KI-Modells 104 zur Regelung der Biomasse-Heizanlage 1 an die Steuereinrichtung 100 der Biomasse-Heizanlage 1 von der zentralen Recheneinheit 190 über das Netzwerk 199;
  • Regeln der Biomasse-Heizanlage 1 durch die Steuereinrichtung 100 mittels des übertragenen KI-Modells 104.

The following steps are also (optionally) provided:

• (Further) aggregating the sensor data transmitted from the majority of the biomass heating systems 1 in the central processing unit, with which the respective sensor data of the biomass heating systems 1 are summarized;
• Machine learning using the machine learning unit 101 using the aggregated sensor data, resulting in a parameterized AI model 104;
  • Transmission of the parameterized AI model 104 for controlling the biomass heating system 1 to the control device 100 of the biomass heating system 1 from the central processing unit 190 via the network 199;
  • Regulation of the biomass heating system 1 by the control device 100 using the transmitted AI model 104.

• Erfassen von chemischen und/oder physikalischen Größen der Biomassen-Heizanlage (1) durch eine zumindest einen Sensor 111-117, 582, 592 (oder eine jeweilige Mehrzahl von Sensoren) der Biomasse-Heizanlagen 1, welche mit der Steuereinrichtung 100 kommunikativ verbunden ist (sind);
• Ansteuern zumindest eines Aktors (oder einer Mehrzahl von Aktoren der Biomasse-Heizanlage 1), welcher mit der Steuereinrichtung 100 kommunikativ verbunden ist (sind) basierend auf den erfassten Größen und dem übertragenen KI-Modell.

The regulation step can preferably have the following steps:

• Detection of chemical and/or physical parameters of the biomass heating system (1) by at least one sensor 111-117, 582, 592 (or a respective plurality of sensors) of the biomass heating system 1, which is communicatively connected to the control device 100 ( are);
• Controlling at least one actuator (or a plurality of actuators of the biomass heating system 1), which is (are) communicatively connected to the control device 100 based on the detected variables and the transmitted AI model.

The above is also in relation to the 22 discussed in more detail.

In order to improve the quality of the prediction or use of the (control) model for the biomass heating system 1 and to optimize machine learning solutions for more complex applications or to enable these applications, sufficient amount of training data makes sense.

The principle applies that the larger the amount of training data, the greater the prediction accuracy of the model. In this respect, it is an advantage of the central approach presented above that a large number of sensor data from a large number of biomass heating systems is collected and aggregated. This means that the database for machine learning is larger and the result of the learning is more precise.

Although small machine learning models can also be trained (locally) with smaller (local) amounts of data, the computing power required to learn larger models in neural networks also increases exponentially with the number of parameters and recorded variables.

In this case, it is a problem to make a correspondingly high computing power available in a commercially viable manner using the computing units that are usually present locally.

To solve this problem, it is possible with the approach explained here to transfer the workload for machine learning from the local control device 100 to a central (and preferably specialized) server 190, which also has the data from a large number of biomass heating systems can. As a result, not only is the database larger, but a significantly greater computing power (for example by means of a server optimized for computing large amounts of data) can also be made available in a resource-saving manner.

In addition, with the central approach with supervised machine learning, it is advantageously possible for a specialist from the server operator (manufacturer) to monitor the learning process without the local operators of the biomass heating systems 1 having to make the effort for this or without this knowledge of machine learning ( and the corresponding operation) must have.

In the central approach, learning can also be tried out or tested before the parameterized AI model is distributed to the biomass heating systems. In other words, only an AI model approved by the manufacturer, which is therefore tested, can be used in the customer's biomass heating system 1, which significantly minimizes the usual risks of using an AI.

20 shows a schematic diagram representing a (example) model for a neuron, which is a possible basis of machine learning. 21 shows a schematic diagram representing a three-layer neural network obtained by summarizing in 21 neurons shown is formed. With the help of 20 and 21 the parameterization or the training of the AI model and also the use of the AI model of the biomass heating system 1 with its basics is explained in more detail.

Neural networks are preferred as learning models for supervised learning and unsupervised learning (or as approximation algorithms for value functions in reinforcement learning).

Specifically, the neural network is realized by, for example, an arithmetic unit, an arithmetic device, and a memory, which are a basic concept of a neuron such as in FIG 20 shown, reproduce mathematically.

As in 20 shown, the neurons serve to provide an output (result) y for multiple inputs x ( 20 represents the three inputs x1 to x3 as a possible example). Each input x(x1,x2,x3) is multiplied by a weight w(w1,w2,w3) equal to input x. With this operation, the neurons output results y given by: $$y=f_k\left({\displaystyle {\sum }_{i=1}^n{x}_i{w}_i-\theta }\right)$$

where 0 is the bias and fk is the activation function. Note that the input x, the result y, and the weight w are all vectors.

An exemplary three-layer neural network constructed by computationally meshing in 20 shown neurons is formed below with reference 21 described.

Multiple inputs x (the three inputs x1 to x3 are presented herein as an example only) are input from the left side of the neural network, and the results y (the three results y1 to y3 are presented herein as an example only) are displayed issued by the right side of this network.

Further, the inputs x1, x2 and x3 are multiplied by a weight corresponding to each of three neurons N11 to N13, and these are then input to the neurons. The weights used to multiply these inputs are collectively referred to herein as W1.

The neurons N11 to N13 output z11 to z13, respectively. Regarding 21 z11 to z13 are collectively referred to as feature vectors Z1 and can be regarded as vectors obtained by extracting the feature amounts from input vectors. The feature vectors Z1 are defined between the weights W1 and W2. Z11 to Z13 are multiplied by a weight corresponding to each of the two neurons N21 and N22 and then input to the neurons. The weights used to multiply these feature vectors are collectively referred to herein as W2.

Neurons N21 and N22 output z21 and z22, respectively. Referring to FIG. 3, z21 and z22 are collectively referred to as feature vectors z2. The feature vectors Z2 are defined between the weights W2 and W3. z21 and z22 are multiplied by a weight corresponding to each of three neutrons N31 to N33 and input. The to multiply these feature vectors The weights used are collectively referred to herein as W3.

Finally, the neurons N31 to N33 output results y1 to y3, respectively. In summary, a neural network processes a plurality of input data and outputs a calculated plurality of output data. At its core, a neural network provides an adaptive calculation function for processing input data.

For the present application to a biomass heating system 1 , this calculation function is an AI model which processes input data from sensors in the biomass heating system 1 and outputs corresponding output data for controlling the actuators in the biomass heating system 1 .

• einen Lernmodus, und
• einen Wert-Vorhersagemodus, der im Kontext der vorliegenden Biomasse-Heizanlage 1 als Regelmodus bezeichnet wird.

The use of the neural network includes two modes:

• a learning mode, and
• a value prediction mode, which is referred to as the control mode in the context of the present biomass heating system 1 .

For example, the weight W is learned (or adapted) using a learning data set in the learning mode, and an output or control variable is determined as an output in the prediction mode or control mode using the weight W.

Although reference was made above to "prediction" or "regulation" for the sake of simplicity, a variety of tasks such as recognition, classification and conclusion or adjustment of control parameters are of course possible with such a neural network. However, since ultimately all of these tasks in turn contain a setting of the actuators of the biomass heating system 1 based on recorded variables, these are collectively referred to here as “rules”. As a recorded size come here for a regulation not only Sizes of the biomass heating system 1 into consideration, but also detected sizes / values from outside the biomass heating system, such as weather data.

Variables or sensor data that are called up while the biomass heating system 1 is being operated with its control device 100 in the control mode can also be learned at the same time and directly, and can thus be reflected in the subsequent action (this is also referred to as "online Learning", whereby learning mode and control mode coincide in this variant).

Alternatively, a set of pre-collected data can be used to perform "collective learning" and subsequently run the control mode using these quantities (this is referred to as collective or batch learning and referred to herein with reference to 22 discussed in more detail). For example, as an intermediate approach, the learning mode can be interposed each time a certain amount of data has been accumulated.

Incidentally, the weights W1 to W3 can be learned by the error feedback method. Information from errors enters on the right and flows to the left. The error feedback method is used to adjust (learn) each weight to reduce the difference between the true output y (teacher) and the output y when input x is input. This can also be referred to as training or parameterizing the model.

Such a neural network can also have more than three layers (this is called deep learning). Likewise, the number of neurons can be predefined in an application-specific manner. Furthermore, the type of networking of the neurons can deviate from the example above.

In this respect, the control device 100 with the machine learning unit 101 for machine learning can have a neural network that is used as a value function or AI model.

The neural network described above can thus form a basis for machine learning, which is described below in the context of the biomass heating system 1 explained herein.

22 10 shows a block diagram representing a basic concept of a data flow or a process of machine learning of a biomass heating system (ie the learning mode of the AI model 104) and the associated execution of the learned model (ie the control mode of the AI model 104). This basic concept can be applied to the regulation of the biomass heating system 1, with the following figures 26 ff. Examples of such applications. the figures 23 and 24 show additions too 22 .

In principle, parameters of an AI model 104 are changed or optimized by an iterative learning process in the learning mode. For example, weights of a neural network are changed. However, parameters relating to the basic setting of the AI model can also be changed in learning mode, for example; for example, the number of neurons in a neural network could be changed.

Furthermore, in principle, a trained AI model with its learned (fixed) parameters is used in the control mode in order to calculate specific output values from input values.

In step S20, sensor data or the variables detected by the sensors # are made available. This can be done using currently recorded data (with online learning) or stored sensor data can be used as a starting point (e.g. with batch learning). The sensor data are preferably stored in a memory 105, 105a of the control device 100 or central computing device 190. In this case, this sensor data can be recorded, for example, as respective data sets for a physical and/or chemical variable over a predetermined period of time and stored in memory 105, 105a.

In this step S20, this data can also be prepared for training as the first part of the training data. On the one hand, this can be done automatically, for example, reference data can be stored automatically with the sensor data, or these training data can be prepared manually from the sensor data by combining them with a so-called “learning input”. The sensor data can be provided by a user, for example, with the desired results or the target specifications as learning input (this addition is optional). For example, a user can supplement sensor data for training with his knowledge of the origin of the data. An example of this is the detection or classification of fuel. The recorded sensor data (e.g. temperature, lambda values, etc.) from various combustion processes (with different fuels) can be supplemented by the user by specifying the label of the fuel. In this case, for example, a number of sensor data can be supplemented with the label “pellets”, and a further number of sensor data can be supplemented manually with the label “wood chips”. This can also further optimize the learning process of the AI model 104 .

For example, data sets with sensor data relating to the use of fuel identification can be provided by a user with supplementary data relating to the fuel identification. A user can thus provide the sensor data with result data (for example, it is specified for a set of sensor data that these measurements were made with pellets as fuel), which are then made available together as the first part of the training data for the next step S21.

At step S21, data filtering takes place. The data filtering is used for the targeted selection and preparation of sensor data and/or the first part of the training data for learning/training (step S23) the model. This data filtering can also be referred to as feature extraction.

In the case of data filtering, filter techniques can preferably be used which reduce the amount of data for learning/training. In addition, (optionally) filter techniques can also be used, which remove incorrectly recorded data or data that is not well suited for learning/training.

• Interpolation; beispielsweise das Ersetzen von unerwünschten 0-Werten durch Mittelwerte des Signals (Interpolationsfilter);
• Glättung; beispielsweise durch einen Median- oder Mittelwertfilter;
• Plausibilitätsfilterung; beispielsweise das Entfernen von zu großen oder zu kleinen Messwerten; Beispielsweise wäre es nicht plausibel, wenn ein O2-Sensor einen Sauerstoffwert von 30% angeben würde (Schwellwertfilter);
• Selektion; Auswahl der zum Lernen bestimmten Daten aus den insgesamt vorhandenen (live-gemessen oder gespeichert) Sensordaten (Selektionsfilter);
• Anwendung fortgeschrittener Filtertechniken, wie beispielsweise HOG (Histogram of oriented gradients, vgl. US 4 567 610 A ) oder SIFT (Scaleinvariant feature transform).
• Wavelet-Filterung, beispielsweise Haar-Wavelets.

For example, step S21 extracts data from the stored data from step S20 using at least one filter. Such a filter can do the following, for example:

• Interpolation; for example, replacing unwanted 0 values with mean values of the signal (interpolation filter);
• Smoothing; for example through a median or mean filter;
• plausibility filtering; for example, removing readings that are too large or too small; For example, it would not be plausible if an O2 sensor indicated an oxygen value of 30% (threshold filter);
• Selection; Selection of the data intended for learning from the total available (live measured or stored) sensor data (selection filter);
• Application of advanced filtering techniques such as HOG (Histogram of oriented gradients, cf. U.S. 4,567,610 A ) or SIFT (scaleinvariant feature transform).
• Wavelet filtering, such as Haar wavelets.

The selection using a selection filter is probably the most important point of the data filtering of step S21. When filtering data by selection, the sensor data that is to be used for machine learning is always selected first. For example, it specifies that sensor data is only used by certain sensors. Consequently, the AI model with predefined sensor data work. In addition, the exact time period of the data can also be specified during the selection.

In summary, in step S21 data can be filtered by selecting sensor data, which can be selected according to the type of sensor data and/or the time period of the sensor data.

Furthermore, so-called “labels” are made available in step S22 if monitored machine learning is used. In the case of unsupervised machine learning, step S22 is unnecessary. As discussed earlier, each label is the name of a particular concept, category, or class of data that the model is learning to recognize. A label can be the type of fuel, for example, or the “combustion chamber temperature” label, whereby all sensor data recorded by combustion chamber temperature sensor 117 are referred to as sensor data with the “combustion chamber temperature” label.

In step S22, these labels are combined with the filtered sensor data or the first part of the training data, with a corresponding data matrix of labels and filtered and prepared sensor data being created. The labels provided the second part of the training data.

This data matrix thus contains the complete training data for the AI model 104.

Step S23 relates to the actual learning/training of the model. In this step, for example, a predefined/preconfigured neural network is trained, which is thus in the learning mode. The number of neurons (which are also referred to as “nodes” in programming) and the number of layers in the neural network are usually predetermined or can be adapted to the application, depending on the application of the model. For example, 600 neurons can be used in 3 layers. However, if, for example, the complexity of the application is greater and the prediction accuracy of the AI model 104 is not satisfactory, the number of neurons can be increased. In essence, the AI model learns from its input how "good" the previous calculation was and changes its parameterization accordingly to optimize its prediction accuracy. After a number of iterative learning processes, the prediction accuracy of the AI model 104 has usually improved significantly, whereby the parameterization of the AI model 104 has been optimized. This parameterization, for example the weights of the neural network, can be processed and stored as parameter data 106 (cf. 24 as an example). The result of the learning process is therefore largely contained in the parameter data 106 . With previously known framework conditions relating to the AI model 104, in the simplest case it is sufficient to only transfer the parameter data 106 from the training model to the AI model for control (for the control mode) in order to effectively pass on the result of the learning process in the learning mode.

As a result of step S23, the training of the model, a parameterized model is created, which is stored as such in step S24. This model is preferably a parameterized neural network with its weights. In this case, the basic parameters of the neural network are basically predefined, but the weightings or weightings of the neural network change in step S23. The weightings of the neural network are thus variable parameters of the model, which are changed application-specifically with learning.

24 shows an example of an excerpt from parameter data that reflect the weights of a parameterized neural network. A number stands for a weight of the neural network. The position of the number in the data also indicates the position of the weight in the neural network. It is therefore sufficient to store the parameter data, for example as an array or as a string in the manner shown, in order to adequately define the variable parameters of the neural network. Otherwise, parameter data can also contain other parameters, for example the number of layers required.

In step S25 of 22 an evaluation of the model is carried out. This can either be done automatically (for example, a desired level of accuracy is not achieved) or a manual evaluation is carried out by a user.

Steps S23, S24, S25 and optionally S21 are carried out iteratively (in at least two runs, more likely in dozens or even hundreds of runs or so-called epochs), with the learning/training of the model being repeated in order to improve the accuracy/prediction quality of the to improve the model. After evaluating step S25, this iteration usually leads back to retraining step S23. However, the evaluation of the model can also show that, for example, sensor data from another sensor is required or that filter parameters have to be adjusted. In this respect, the iteration can also lead back to step S21 in exceptional cases.

In this regard, the Figures 23a and 23b Examples of learning curves as a result of the iterative learning/training of the model, which were obtained during training of the model for fuel identification of the biomass heating system with data sets from this system. The "epoch" horizontal axis refers to the number of iterations, with one (1) "epoch" denoting one (1) pass in training the model. The vertical axis "cross entropy" generally explains the accuracy of the model; the smaller the value on this axis, the better or more accurate it is.

In information theory and mathematical statistics, the cross entropy is a measure of the quality of a model for a probability distribution. Minimizing the cross-entropy with respect to the model parameters is equivalent to maximizing the log-likelyhood function. The solid line shows the cross entropy of training and the dashed line the cross entropy of validation.

The effect of repeatedly training the model can be seen in this figure: the model's prediction accuracy increases with the number of training runs. It should be noted that the larger the number of available training data for each iteration (for each epoch), the more effective the training.

If it is found when performing the iterations that the AI model is sufficiently accurate, the training is terminated. As a result of the iterative process of 22 , which is in the upper third of the 22 is shown, there is a parameterized AI model for a specific application.

The trained parameters, ie parameter data 106, or the entire AI model 104 is/are then transferred to the control device 100 of the biomass heating system 1. This can then carry out an optimized regulation of the biomass heating system 1 . In detail:
At step S30, sensor data is provided. These can either be provided directly by sensors during operation by detecting at least one physical and/or chemical parameter, or detected sensor data can be retrieved from memory 105, 105a of control device 100 (in which these are temporarily stored).

The sensor data provided are again subjected to data filtering in the following step S31. This step S31 usually corresponds to step S21, to which reference is made. This creates a data vector or feature vector with the filtered data for the AI model 104. This data vector is supplied to the AI model 104 in the control mode. The weights of the model usually remain unchanged. A model result, which is used in the following step S40, is calculated with the aid of the AI model.

In step S40, the model result is converted into concrete instructions for controlling at least one actuator of the biomass heating system 1. If the model result indicates, for example, that a certain type of fuel has been recognized, corresponding presettings (for example a position of a valve 52 of the recirculation device 5) are retrieved.

With these instructions, at least one actuator of the biomass heating system 1 is then set/controlled in step S41.

Due to the setting of the at least one actuator of the biomass heating system 1, at least one physical or chemical variable in the biomass heating system 1 can now also change.

For example, the temperature in the combustion chamber 24 can change due to the change in a position of a valve 52 of the recirculation device 5 . In this respect, sensor data relating to the combustion chamber temperature will in turn change. A control circuit is thus present, with which the biomass heating system 1 can be controlled by means of the AI model.

In order to better discuss the regulation of the biomass heating system 1, reference is made below to FIG 25 a combustion operation or an operating method of the biomass heating system 1 is explained.

After the combustion operation has started, usually by switching on the biomass heating system 1 by a user or by an external automatic system, the combustion process can first be prepared in the optional step S50.

When preparing step S50, the biomass heating system can be mechanically and electronically initialized. For example, the operating system of the control device 100 starts up, a self-test of the electronics is carried out and/or the rotary grate elements 252, 253, 254 are rotated (opened) by a predetermined angle in order to remove any deposits on the grate and the mechanics forward to test a combustion process. In such a mechanical test of the rotary grate 25, the rotary (transmitter) sensors can be used to check whether activation of the motors 231 of the rotary mechanism leads to the desired result, or whether something is blocked. Furthermore, the mechanical boiler cleaning (via tubulators), the ash removal and the optional electrostatic precipitator cleaning for a predefined Time (e.g. 30 seconds) are operated. The airways of the boiler 11 can also be flushed. To do this, the biomass heating system is flushed with air by opening the primary air and secondary air valves. Then the air slides are closed and the flue gas recirculation line is purged.

In the next step S52, the combustion chamber 24 is filled with fuel. The fuel is conveyed to the rotary grate 25 via the fuel supply 6 until a predetermined height of the fuel bed is reached. To do this, the fuel bed height is measured with the fuel bed height sensor 116 . The fuel bed level sensor 116 is, for example, a mechanical level flap 86 with a rotation angle sensor.

Next, the fuel is ignited in step S52. This can also be referred to as the ignition phase. Energy is supplied to the fuel via the ignition device 201 until it burns. Furthermore, when the fuel is ignited, the valves or valve positions can be set in such a way that they promote the ignition of the fuel. Incidentally, with such an ignition, the blower 15 is also activated in order to generate a corresponding negative pressure in the combustion chamber 24 . The primary air and secondary valves can be set to predefined values (e.g.: 60% and 15%) and a predefined negative pressure in the combustion chamber (e.g. 75 Pa) is regulated.

If the biomass heating system 1 now reaches a predetermined combustion chamber temperature (for example 50° C.) and/or a predetermined lambda (for example 17%), the biomass heating system 1 goes to step S53, the stabilization of the combustion. This step, also referred to as the stabilization phase, further promotes ignition of the fuel bed. The positions of the air valves 52, the function of the blower 15 and also the fuel supply are adjusted accordingly. The boiler 11 and also the combustion chamber 24 should continue to heat up. Preferably, the combustion process should gradually transition to a stationary state in which there is an equilibrium from a thermodynamic point of view. If the combustion temperature increases up to a predetermined value, for example 400°C, step S53 is completed.

Accordingly, the process proceeds to step S54: the stabilized combustion and the actual heating operation. In this step S54, the power output or the combustion intensity is controlled by means of the fuel supply 6, the blower 15, the position of the valves 52, and other actuators based on the sensor data from sensors of the biomass heating system 1, for example based on the combustion chamber temperature, the Lambda value and/or the boiler (water or medium) temperature.

• Unterdruckregelung

In step S54, at least one of the following control concepts of the biomass heating system 1 can optionally be used:

• vacuum control

• O2-Regelung bzw. Lambda-Regelung

A sufficiently high negative pressure in the combustion chamber must always be maintained so that (i) the primary and secondary combustion air can be sucked in from the outside and the recirculated flue gas can be sucked in from the flue gas duct and (ii) no overpressure is created in which flue gases can escape via the inlet air openings and any leaks in the combustion chamber can escape from the installation room.

• O2 control or lambda control

In order to ensure low-emission combustion, the boiler must be operated with a certain amount of excess air. The so-called combustion air ratio (= supplied air quantity divided by the minimum air quantity required for combustion of the fuel under stoichiometric conditions) must therefore be greater than 1. However, it should also be kept as low as possible, since the combustion efficiency of the boiler decreases with increasing excess air.

• Leistungsregelung

Biomesse boilers are usually operated with combustion air ratios between 1.5 and 2.3 (corresponds to the oxygen content in the dry flue gas between 7 and 12 vol%). Due to the particularly efficient combustion chamber and combustion air injection concept optimized with CFD simulations, significantly lower combustion air ratios can be achieved with the present biomass heating system 1, which are typically between 1.3 and 1.75 (or 5 and 9 vol% O2 im dry flue gas). Such a control can be significantly optimized using machine learning.

• power control

• Brennstoffflexible Leistungsregelung

The task of the power control is to set the actuators of the boiler 11 in such a way that the amount of heat generated in the boiler corresponds to the amount of heat requested by the consumer. For this purpose, a target value for the boiler temperature (=flow temperature) is defined (e.g.: 85°C). If the heat consumption by the consumers increases, the flow temperature falls while the boiler output remains the same. If the heat demand falls, the flow temperature rises. The boiler output is thus adjusted in such a way that this target flow temperature can be kept as constant as possible. This is done by adjusting the fuel supply and adjusting the primary combustion air fed into the fuel bed. The setting values for these parameters depend heavily on the fuel quality (lumpiness, water content, calorific value, energy density). Depending on the calculated required boiler output (in % of the maximum output) and based on the default values for the combustion chamber vacuum at maximum and minimum output, the target value for the combustion chamber vacuum is calculated via a linear comparison between output and vacuum. The speed of the blower 15 can, for example, be continuously adjusted via a PID controller so that the measured negative pressure in the combustion chamber 24 corresponds to the setpoint. Since the negative pressure is constantly changing due to fluctuations in the combustion process that are typical for biomass, the actual value of the negative pressure oscillates to a small extent around the setpoint. Such a control can be significantly optimized using machine learning.

• Fuel-flexible power control

In addition, a fuel-flexible output control is used in the present biomass heating system 1 .

If the combustion output is increased, more combustion air is introduced into the grate area due to the higher negative pressure in the combustion chamber and the higher opening of the primary air slide, which leads to faster burn-off of the fuel bed. As a result, the height of the fuel bed decreases with constant fuel feed. The speed of this height loss also depends on the bulk density, the lumpiness and the water content of the fuel.

In conventional control strategies, a load change is reacted to with a change in the fuel feed via the fuel supply cycle, with the adaptation usually occurring linearly between the setting values for the maximum and minimum power. As a result, it is not possible to react with sufficient accuracy to changes in fuel quality (bulk density, water content).

In the present biomass heating system 1, on the other hand, the fuel insert is adjusted depending on the fuel, taking into account the fuel quality and the power requirement. The clocking of the fuel feeder can be changed in such a way that the fuel bed height measured by the fuel bed sensor remains in a predetermined height range (separately for each type of fuel).

The adaptation of the fuel injection cycle takes place step by step in defined time intervals (e.g.: 120 seconds). The break times are primarily lengthened or shortened. If the cycle time (total of operating time and pause time) in low-load operation exceeds a specific value, the operating time of the slide-in module is also reduced. This results in a aliquot shortened pause time for the same running time of the fuel supply per minute (= the same fuel mass flow introduced). This avoids that too long pause times occur between two racks, which would negatively affect the responsiveness of the control.

Changes in the load requirement and the associated changes in the primary air supply only have a delayed effect (by a few minutes) in the form of a decrease or increase in the height of the fuel bed. Therefore, with this control concept, the fuel feed timing is changed in advance when the gradient of the boiler temperature change over time (indicator of which is a very pronounced increase or decrease in the load requirement) reaches a certain level. In this case, there is also a greater percentage change in the fuel feed (specifically the pause time).

• Brennstoffabhängige Maximallastanpassung

Such a control can also be significantly optimized using machine learning.

• Fuel-dependent maximum load adjustment

With conventional controls, the settings of the actuators for the minimum and maximum power for different fuel qualities have to be specified manually by the user (by manually selecting a fuel). If this were not done, then it could be the case, for example, that the combustion of a fuel with a low water content is controlled with the settings that apply to a fuel with a high water content. In this case, a higher fuel conversion and thus an excessively high output would be achieved. This results in excessive thermal combustion chamber loads, which reduce the service life of the system (increase material loads) and can cause increased emissions.

In the present biomass heating system, the specification for the position of the primary air valve and the combustion chamber vacuum at maximum output can be made for a wood chip fuel of the usual quality (design fuel), i.e. with a water content in the upper third of the permitted fuel spectrum. A corresponding fuel detection is using the 26 explained in more detail.

Step S54 is ended when, for example, sufficient heat output has been made available.

Then, in step S55, the combustion chamber 24 and in particular the rotary grate 25 are burned out. The fuel supply is stopped and the combustion chamber temperature drops. It burns the remains of the fuel on the rotary grate 25. For this purpose, for example, the positions of the valves 52, the fan 15 can be adjusted accordingly. At the end of the burnout, the rotary grate 25 is cleaned by turning or opening the rotary grate elements accordingly.

After the end of step S55, the method can be deactivated (END), or after some time the method can go back to step S50, with which a new heating cycle begins.

(fuel detection)

26 shows a schematic diagram of an exemplary application of machine learning to identify or classify the fuel of the biomass heating system 1. The Figures 27a , 27b and 27c show corresponding sensor data from sensors used for this application (for training and for prediction). The abbreviation "HK" (for wood chips) designates sensor data from the detection of wood chips, and the abbreviation "PL" (for pellets) designates sensor data from the detection of pellets.

• O2, erfasst durch die Lambda-Sonde 112; und
• Abgastemperatur, erfasst durch den Abgastemperatursensor 111; oder
• Brennkammertemperatur, erfasst durch den Brennkammertemperatursensor 117

At least 2 of the following variables are used as sensor data for the AI model 104:

• O2 sensed by lambda probe 112; and
• exhaust gas temperature detected by the exhaust gas temperature sensor 111; or
• Combustion chamber temperature, detected by the combustion chamber temperature sensor 117

These variables can preferably be detected in a period of time that is defined from the ignition (i.e. from the activation of the ignition device 201) until a threshold value of preferably 16% O2 is reached. Alternatively, a longer period can be used.

Sensor data, which came from exemplary measurements with a biomass heating system 1, are in Figures 27a , 27b and 27c shown. All temperature charts of Figures 27a , 27b and 27c show the same period and result for pellets and wood chips from a combustion process after ignition in the biomass heating system 1.

The Figure 27a purely by way of example, the course of a temperature of the exhaust gas after the ignition of wood chips (HK) and, in comparison, of pellets (PL) (cf. horizontal axis, ignition takes place at time 0) in a cut-out period. The exhaust gas temperatures were recorded with the exhaust gas temperature sensor 111 in each case. Both temperature diagrams of wood chips and pellets show the same period after ignition (the extension of the horizontal axis was omitted in the temperature diagram for pellets for the sake of clarity)

The temperature curves of both fuels behave like this Figure 27a in the same (example) recording period from approx. 600 seconds to approx. 1000 seconds.

the Figure 27b shows purely as an example for the same period as Figure 27a comparing the corresponding oxygen content curves, which were determined with the lambda probe 112, for pellets and wood chips. It can be seen here that the corresponding temperature curves are quite similar.

the Figure 27c again shows purely by way of example for the same period as Figure 27a comparing the corresponding combustion chamber temperature curves, which were determined with the combustion chamber temperature probe 117, for pellets and wood chips. Despite the similar course of the curve, clear differences in temperature can be seen here. For example, the temperature curve for the wood chips (HK) ends at around 900°C, while the temperature curve for the pellets (PL) ends at around 1150°C

With these examples of Figures 27a , 27b and 27c it is illustrated again what can be understood by sensor data. In the present example, it is sensor data that is recorded over a period of time and reflects a physical or chemical variable. For example, the sensor data of the exhaust gas temperature sensor 111 includes the temperature of the exhaust gas over time.

Depending on the application, the time period of the sensor data can advantageously be limited to a time period that is particularly characteristic of the application. It has been shown, for example, that the type of fuel, especially after it has been ignited, leads to sensor data that is characteristic and thus easily distinguishable for an AI.

Referring to 26 This sensor data selected according to time and source (sensor) can be used to train the AI model 104 and to execute the AI model 104 during operation of the boiler 11 in order to classify the fuel.

Testing by the present inventor has shown that training with sensor data from the combustion of different fuels and also with sensor data from (the same) fuels of different quality leads to an AI model 104 which makes good predictions about the type of fuel and its quality can.

Consequently, this classification can not only relate to the general type of fuel, but also roughly divide the fuel into its quality (especially its water content).

In this regard, it should be noted that during training, a classification model is assigned the 26 sensor data shown (or the oxygen content of the exhaust gas of the biomass heating system and at least one of the two in 26 other sizes shown) and also the associated labels. These labels are in 26 not shown separately for the sake of clarity, since these are of course already included in the context of the respective recorded variable.

Given enough training data (often hundreds or thousands of sensor data or even datasets per label), a classification model can learn to predict whether new data belongs to one of the classes it was trained on. This prediction process is called inference.

• Label 1: 80%
• Label 2: 15%
• Label 3: 5%

If one then provides new or current sensor data as input for the model when using or predicting, the trained model outputs probabilities regarding the sensor data, these being output for each label for which it was trained. An example output from the trained model or AI could be as follows:

• Label 1: 80%
• Label 2: 15%
• Label 3: 5%

Each row of this example corresponds to a label of the training data (ie the second part of the training data). The classification with the highest probability can be determined accordingly by means of this output of the probabilities. In this case the label 1 would be classified or recognized. For example, label 1 could be pellets, label 2 for wood chips and label 3 for elephant grass, with which pellets would be recognized as fuel in the present example.

In other words, the prediction result of the AI model 104, which is to be statistically classified regularly, is evaluated by the control device in such a way that the result with the greatest prediction probability is used as the final result of the model 104.

Labels called “fuel quality 1 . . . 10” could now be used as further labels relating to the quality of the fuel, which is an indicator of the quality of the fuel on a scale of 1 to 10 (from poor to very good), for example. In this way, the AI model 104 could be provided with further labels in the second part of the training data in order to classify the fuel even more precisely. The same also applies to the numerical values output by the AI model, which are also sorted by label (e.g. oxygen content 16.1% (vol%), oxygen content 16.2%, oxygen content 16.3"...) with their probabilities be issued.

Such labels are used in the examples of 28 ff. are not mentioned separately, as these naturally result from the type of sensor data or from the respective information. However, these are then correspondingly in the 28 ff.

One of the properties of the present biomass heating system 1 that is to be emphasized is that it should be able to be fired in a hybrid manner. A wide variety of carbon-based solid fuels can be used, such as wood chips or pellets, but also reed grass (miscanthus or elephant grass) or olive shells.

The aforementioned fuels usually have very different burning behavior. Due to their natural origin, such biogenic fuels do not usually burn in the same way, although there are now a number of standards relating to fuels, such as pellets. So (even with standardized fuels) the length, the dust content, the type of wood, etc. of the fuel vary. In addition, the consistency or the water content of the fuel can vary greatly. This can also be caused, for example, by improper storage or non-quality-assured production of the fuel.

However, the type of fuel determines the control parameters of the biomass heating system 1 significantly, which in turn strongly influences the combustion process. For example, the detection of the fuel influences or determines the operation of the actuators of the boiler 11, for example the control parameters of the air valves, the material supply quantity of the fuel or the cleaning process of the boiler 11 (e.g. cleaning intervals and intensity). Thus, incorrect identification of the fuel usually leads to a reduction in the efficiency of the biomass heating system 1 to the point of malfunctions in the biomass heating system 1 during operation, for example due to increased slagging.

Incorrect material detection can also lead to a delay in the start-up of the boiler 11 or to a long start-up time, a rapid drop in performance and a loss of performance.

However, conventional biomass boilers 1 have a major problem in correctly classifying the type (and quality) of the fuel, since they usually rely purely on spot measurements from the lambda sensor. The type of fuel in conventional biomass heating systems 1 is regularly determined only by means of the oxygen content of the exhaust gas/flue gas and associated threshold values in steady-state operation, for example after 30 minutes or more. Experiments have shown that this rudimentary implementation of fuel identification has a comparatively high error probability and, moreover, can fail completely with more than two fuels for classification due to the lack of ability to allocate the oxygen measurement to a specific fuel.

In addition, with this type of fuel detection, the fuel is determined with a delay, which means that the boiler 11 is already started up with permanently set parameters, and the control of the boiler 11 is thus adapted to the type of the fuel can only take place at a late point in time in the combustion process.

Using the AI model 104 to classify the fuel of the biomass heating system 1, which with the context 26 explained sensor data works, the detection accuracy or classification accuracy of the fuel can be significantly improved.

With the result of the classification of the fuel, the type of fuel can be determined on the one hand, and optionally also the quality of the fuel. This in turn determines various basic settings of the biomass heating system 1 . For example, the various basic settings (for example basic valve positions of the valves 52, fan working ranges of the fan 15, etc.) can be pre-stored in the control device 100 for each type (and quality) of fuel. Once the fuel has been classified, these basic settings are retrieved and applied.

For the figures 28 ff. applies that the above in relation to the Figures 18 to 27 The basics explained apply and these are not repeated. With regard to the sensor data or physical or chemical variables mentioned for the AI model 104, the following applies: Figures 1 to 27 mentioned sensors are used for detection. The correspondingly made available sensor data are also used for training and for executing the AI model 104 .

28 shows a method for material identification -/classification of fuel of the biomass heating system 1 with an AI model 104;

The lambda of the exhaust gas, the combustion chamber temperature, the flue gas temperature and, in addition to increasing the accuracy of the model, the time from ignition to reaching a predefined combustion chamber temperature and the time from ignition to reaching a predetermined fuel bed height can be recorded.

29 shows a method for fuel bed height control of the biomass heating system 1 of FIG Figures 1 to 24 with an AI model 104.

An AI model 104 for fuel bed height control is provided with a preset target fuel level and sensor data of at least one of the following variables: Lambda or O2, combustion chamber temperature, flue gas temperature, ignition time (i.e. the time from initiation of ignition to to reach a predetermined temperature in the combustor) and fill time (i.e. the time from the start of fuel injection into an empty combustor until a predetermined fuel bed height is reached).

Each of the above variables or the resulting sensor data can be in one of the 25 mentioned phases or also in several of them. The phase of (regulated) combustion (step S54) has proven to be particularly advantageous as a time period for the sensor data, in which the height of the fuel bed should ideally be kept quite constant at an "ideal value" in order to keep the amount of fuel available and thus the fuel energy constant to keep.

The result of the execution of the AI model 104 in training or "learning" (ie in the training mode, which is referred to as learning mode here) and in ongoing operation (ie in the control mode) can preferably be supplied as a control parameter to a control algorithm 107, which Motor 66 drives the fuel supply 66 and is informed by the height sensor 116 of the height of the fuel bed. Such a control algorithm 107 can be a conventional P, PI, PID, PD algorithm in which, however (in contrast to the prior art) one of the control parameters is at least influenced by the (further) control parameter of the AI model 104 . For example, the result of the AI model 104 may be the P, I, or D term of the control algorithm, or at least a portion thereof (e.g., a P term of the AI model in a predefined range of values may be added to a predefined offset). Thus, for example, the P component or the I component of a PID algorithm can come from the AI model 104 . In addition all control parameters of the control algorithm 107 can also come from the AI model 104 .

It should be noted that the control algorithm 107 can also be used for all other controls explained here (cf. Figures 18 to 34 ). In this respect, the result of the calculation of the AI model 104 can be used directly to control an actuator etc. (e.g. the boiler 11 can be switched on or off) or the result of the calculation of the AI model 104 can be used as an input variable for a control algorithm 107 serve, which performs control calculations. Alternatively, the result of the calculation of the AI model 104 can also be used to set basic settings (for example control ranges of the air valves 52, cleaning intervals, etc.) of the biomass heating system 1. An example of this would be the result of the classification of the fuel. If a specific fuel is detected, a cleaning interval corresponding to the fuel is set.

Conventional control algorithms, which work without an additional input from an AI model, lead to strong fluctuations in the height of the fuel bed, as these are too coarse, too imprecise and/or not time-adjusted. Using the AI model 104, the biomass heating system 1 can "learn" to better regulate the height of the fuel bed, taking into account the specifics of the system (i.e., taking into account the type of fuel and the specifics of the biomass heating system 1 itself). In particular, the AI can be used to determine what level is required to achieve the desired energy class of the boiler 11 using the parameters from now and from previous data.

30 shows a method for lambda control or for oxygen content control of the exhaust gas of the biomass heating system Figures 1 to 24 with an AI model. The aim of the method is an optimized lambda control of the biomass heating system 1. The lambda value or oxygen (residual) content of the exhaust gas is, as already explained, a decisive factor that influences the quality and efficiency of the combustion process. In this case, with the AI model 104, the Combustion process in the biomass heating system 1 optimized.

Sensor data of at least one of the following variables is made available to an AI model 104 for lambda control: detected lambda or O2, combustion chamber temperature, exhaust gas temperature, negative pressure in the combustion chamber, position or degree of opening of the at least one primary valve 52, position or degree of opening of the at least one secondary valve 52, and fuel bed height. The desired power output of the boiler 11 is also made available to the AI model 104 as a default.

Each of the above variables or the resulting sensor data can be in one of the 25 mentioned phases or also in several of them. The phase of (regulated) combustion (step S54) has proven to be particularly advantageous as a time period for the sensor data for the application for lambda control, since lambda control to an optimum lambda value has the greatest effect in this time period in terms of efficiency and cleanliness of the combustion process.

With AI, the biomass heating system thus calculates which lambda level is optimal to work with the parameters and how the behavior was in the past with these settings, especially with secondary air and with outlet air.

31 shows a method for on/off control of the biomass heating system Figures 1 to 24 with an AI model 104. One of the basic rule decisions when heating a house or similar is the decision to activate or not activate the biomass heating system at the right time.

A biomass heating system 1 that is activated too late, for example during a cold snap, might under certain circumstances not be able to provide sufficient power, with the result that the temperatures in the house could drop uncomfortably.

Conversely, a biomass heating system that is only activated "too much as a precaution" during a brief cold snap could provide excess power that is not required, and thus be inefficient.

Likewise, for example, an occurring heat wave can make the use of the biomass heating system 1 superfluous.

It has been shown that conventional biomass heating systems are either not activated and deactivated with foresight at all, or are activated or deactivated only in a very static or limited framework by means of user interaction.

However, since an AI model can learn from the "experience" of a large number of heating processes and weather events, on/off control of the combustion process of the biomass heating system 1 can also be optimized.

An AI model 104 for lambda control is provided with sensor data from the following variables: a predetermined energy class of the biomass heating system 1 (e.g. 50 or 120 kW, i.e. the capacity of the biomass heating system 1), the current temperature of a heating water buffer ( or to put it more generally: a currently available amount of heat for heating), optionally a specified time to react to a change in the weather, and data on the development of the weather.

The data on the development of the future weather preferably includes the (outside) temperatures over a period of one (1) day or up to three days. In particular, data can be used that contain, for example, the temperatures predicted by the weather service at the location where the biomass heating system 1 is set up every hour. Likewise, the future weather data may include data about future snowfall and/or future rainfall. Indicators about the future power requirement for heating with the biomass heating system 1 can be derived from this weather data. In other words, future weather data is the data retrieved from a weather service as a forecast can be and which allow a prediction of an expected power requirement of the biomass heating system 1.

From a comparison of the (future) weather data with the performance (output) capability (or the energy class) of the biomass heating system 1, as well as the currently available amount of heat energy, a suitably trained AI model can make an optimized forecast of the time at which the ( Combustion in the) biomass heating system 1 is activated or deactivated.

32 shows a method for weather-dependent power control of the biomass heating system Figures 1 to 24 with an AI model.

A conventional power output regulation of a heating system is usually carried out using thermostats on the radiators and an outside temperature sensor, with which the general power requirement of the system is estimated. However, many other weather factors can at least influence the general power requirement of the heating system, for example rain, snow, solar radiation and standstill, wind, etc. Likewise, a power output from a heating system is usually only determined with the current temperature, a forward-looking control of the power output based on a weather forecast not usual.

Conventional output controls for biomass heating systems 1 therefore often have the problem that they often involve high output fluctuations. However, this contradicts the ideal case of a combustion process that is as steady and uniform as possible at the optimal (power) operating point or in the optimal power operating range of the biomass heating system 1.

To solve this problem, an AI model for weather-dependent power control is used here, which on the one hand can work with foresight and on the other hand can also avoid excessive power fluctuations.

AI model 104 receives at least one of the following variables as sensor data: detected lambda or O2, combustion chamber temperature, exhaust gas temperature, negative pressure in the combustion chamber, position or degree of opening of the at least one primary valve 52, position or degree of opening of the at least one secondary valve 52, fuel bed height , as well as water flow into or out of the boiler, a boiler temperature (or, in relation to the two above variables, formulated more generally: a heat quantity sensor relating to the heat quantity given off by the boiler, for example a flow quantity sensor with a temperature sensor and associated calculation of the heat quantity).

Each of the above variables or the resulting sensor data can be in one of the 25 mentioned phases or also in several of them. The phase of (regulated) combustion (step S54) has proven to be particularly advantageous as a time period for the sensor data for use in power control.

The AI model 104 continues to receive the boiler target output as a default.

The AI model 104 also receives at least current weather data, for example the current outside temperature at the location of the biomass heating system 1, and preferably additional weather data, such as the position of the sun.

In addition, the AI model 104 may receive weather forecast data over a predefined future time period, such as in relation to the 31 already explained.

The result of the AI model 104 can be fed to a control algorithm 107, as described above in relation to FIG 29 already explained what is referred to.

In this case, the control algorithm 107 will adjust the output power of the boiler, which in turn is recorded by means of suitable sensors as the actual boiler output. the The actual boiler output is in turn fed to the control algorithm 107 as a control input variable.

Figure 33 shows a method for (now weather-independent) output control or temperature control of the biomass heating system Figures 1 to 24 with an AI model 104.

In this method, the AI model 104 uses at least two of the following variables or the resulting sensor data: Combustion chamber temperature, boiler temperature, amount of heat given off (determined, for example, by means of a heat amount sensor, which determines the flow rate and the temperature of the water flowing through and uses this to determine the amount of heat) , fuel bed height.

Each of the above variables or the resulting sensor data can be in one of the 25 mentioned phases or also in several of them. The phase of (regulated) combustion (step S54) has proven to be particularly advantageous as a time period for the sensor data for use in power control.

The result of the AI model 104 is in turn fed to a control algorithm 197 . This controls towards a boiler temperature, which in turn is detected by the boiler temperature sensor 115, from which in turn a deviation T2-T1 (setpoint temperature T2 minus detected boiler temperature T1) is determined, which in turn is fed to the control algorithm 107.

This can also be used to regulate the boiler output or temperature in an optimized manner.

34 shows a method for primary air control of the biomass heating system Figures 1 to 24 with an AI model. The position or the degree of opening of the at least one primary valve 52 and also of the secondary valve 52 have significant influences on the optimization of the combustion process in the boiler 11 . A regulation can also be used here contribute significantly to optimization using an AI model.

In this method, the AI model 104 uses at least two of the following variables or the resulting sensor data: Combustion chamber temperature, boiler temperature, amount of heat given off (determined, for example, by means of a heat amount sensor, which determines the flow rate and the temperature of the water flowing through and uses this to determine the amount of heat) , fuel bed height.

Each of the above variables or the resulting sensor data can be in one of the 25 mentioned phases or also in several of them. The phase of (regulated) combustion (step S54) has proven to be particularly advantageous as a time period for the sensor data for use in power control.

The position of the at least one secondary valve can be derived directly from the result of the AI model 104 and the position of the at least one primary valve 52. In addition, the actual output of the boiler 11 is recorded and included in a differential control of the position of the at least one primary valve 52 depending on the actual output of the boiler 11 and the position of the primary valve 52 .

For example, the primary air is also reduced if the secondary air is above a specified target value. With the use of the KI model 104, the target value of the primary air for a specific output can be maintained and with the help of the learning process of the KI model 104 from "old data", small adjustments to the system can be optimized from the secondary air by means of the position of the secondary air flap Calculate and adjust the setting parameters of the valves 52 accordingly. In this way, fluctuations in the boiler airflow system can be significantly reduced

(Further embodiments)

In addition to the explained embodiments and aspects, the invention permits further design principles. Thus, individual features of the various embodiments and aspects can also be combined with one another as desired, as long as this is apparent to the person skilled in the art.

The recirculation device 5 is described here with a primary recirculation and a secondary recirculation. However, in its basic configuration, the recirculation device 5 can also only have a primary recirculation and no secondary recirculation. With this basic configuration of the recirculation device, the components required for the secondary recirculation can be omitted completely, for example the recirculation inlet channel divider 532, the secondary recirculation channel 57 and an associated secondary mixing unit 5b, which will be explained later, and the recirculation nozzles 291 can be omitted.

Alternatively, only one primary recirculation can be provided in such a way that the secondary mixing unit 5b and the associated channels are omitted, and the mixture of the primary recirculation is not only fed under the rotary grate 25, but also (e.g. via another channel) to the is supplied to the recirculation nozzles 291 provided in this variant. This variant is mechanically simpler and therefore less expensive, but still has the recirculation nozzles 291 for creating a swirl in the flow in the combustion chamber 24 .

Furthermore, instead of only three rotary grate elements 252, 253 and 254, two, four or more rotary grate elements can also be provided. For example, five rotary grate elements could be arranged with the same symmetry and functionality as with the three rotary grate elements presented. In addition, the rotary grate elements can also be shaped or designed differently from one another. More rotary grate elements have the advantage that the crushing function is increased.

With regard to the specified dimensions, it should be noted that other dimensions or combinations of dimensions can also be provided that deviate from these.

Instead of the convex sides of the rotary grate elements 252 and 254, concave sides of these can also be provided, in which case the sides of the rotary grate element 253 can be shaped in a complementary convex manner. This is functionally almost equivalent.

Fuels other than wood chips or pellets can also be used as fuels in the biomass heating system. In particular, olive skins, elephant grass and biogenic waste can be used as additional fuels.

The biomass heating system disclosed here can also be fired exclusively with one type of fuel, for example only with pellets. The fuel identification can also be used in such a way that the quality (eg the water content) of only one fuel is identified.

The combustion chamber blocks 29 can also be provided without the recirculation nozzles 291 . This can apply in particular to the case in which no secondary recirculation is provided.

The rotational flow or turbulent flow in the combustion chamber 24 can be clockwise or counterclockwise.

The combustion chamber cover 204 can also be provided with an incline in sections, for example in a stepped manner.

The secondary (re)circulation can also only be flown with secondary air or fresh air, and in this respect not recirculate the flue gas, but only supply fresh air.

The secondary air nozzles 291 are not limited to purely cylindrical bores in the combustion chamber bricks 291 . These can also be designed as frustoconical openings or tapered openings.

The specified dimensions and sizes are only to be understood as examples and can be modified.

Here, the recirculation device 5 in the embodiment of 12 with a primary recirculation and a secondary recirculation. However, in its basic configuration, the recirculation device 5 can also only have a primary recirculation and no secondary recirculation. With this basic configuration of the recirculation device, the components required for the secondary recirculation can be omitted completely, for example the recirculation inlet channel divider 532, the secondary recirculation channel 57 and an associated secondary mixing unit 5b, which will be explained, and the recirculation nozzles 291 can be omitted.

Alternatively, only one primary recirculation can be provided in such a way that the secondary mixing unit 5b and the associated channels are omitted, and the mixture of the primary recirculation is not only fed under the rotary grate 25, but also (e.g. via another channel) to the is supplied to the recirculation nozzles 291 provided in this variant. This variant is mechanically simpler and therefore less expensive, and nevertheless has the recirculation nozzles 291 for creating an eddy current or twist in the flow in the combustion chamber 24 .

An air quantity sensor, a vacuum unit, a temperature sensor, an exhaust gas sensor and/or a lambda sensor can also be provided or only at the inlet of the flue gas recirculation device 5 .

The sensors 111 to 117 can be provided in any suitable location for measuring the respective quantities, and as such the locations for these sensors not limited to the locations shown in the figures.

Although a large number of sensors are shown and described in the figures and in the embodiments, only one of these sensors or only some of these sensors can also be used for machine learning in the sense of the applications presented. In other words, not all of the sensors explained are necessary in order to implement machine learning in the sense of the present embodiments and aspects. In particular, the combinations of the individual sensors (and thus the measured physical variables) can be used in the individual application examples 25 ff. are listed.

The position of the sensors 111, 112, 113, 114, 116 and 117 etc. is only shown as an example in the figures and can vary as long as the respective intended operating variable can be detected directly or indirectly (e.g. by corrective calculation).

Likewise, more or fewer sensors or actuators than indicated in the figures can also be provided. For the respective application 26 ff. Only certain sensors are required, with only a part of the 26 ff. specified sensors may be sufficient. However, of course, the described principle applies that the respective AI models can provide better predictions the more input data is used for training and execution. In this respect, the AI models described here are all the more accurate the more sensors with corresponding sensor data are used.

Although a local and a central training of the AI model are described, in principle a mixed form of this can also be used. For example, part of the training data can be obtained on site and the model can be trained with it, and this can then be further trained centrally with training data from other biomass heating systems.

Actuators which are not explicitly mentioned here also come into consideration as actuators which are regulated by the biomass heating system 1 . The term "actor" includes any actor, regardless of whether it directly or indirectly influences the combustion processes in the biomass heating system or not. For example, an indication on a display showing the result of the identification of the fuel (the classification of the fuel) can be understood as an "actor" within the meaning of this application. According to a special understanding, however, the term "actor" of this disclosure can also be understood as just an actor which can influence the combustion processes in the biomass heating system 1 (for example an (air or water) valve, a fan, a heating water pump, a rotary grate servomotor, a motor for cleaning the biomass heating system 1.

Any data that provides information about the future development of the weather (e.g. temperature developments, probability of rain, etc.) can be understood as weather forecast data. Such data is usually provided online by so-called weather services in a predetermined data format, with which it can be processed by the AI model 104 after pre-filtering. Protocols for retrieving the weather forecast data are known to the person skilled in the art and are required for this application in terms of content.

Mathematical models or concepts other than the neural network explained herein can also be used as the AI model.

A computer program, which may also be referred to or described as a program, software, software application, application, module, software module, script, or code, may be written in any form of programming language, including compiled or interpreted languages, or declarative or procedural Languages; and may be employed in any form, including as a specific program, or as a module, component, subprogram, or other suitable entity for use in a computing environment. A program can, but does not have to, use one match file in a file system. A program may be stored in a portion of a file that holds other programs or data, e.g. one or more scripts stored in a tracking language document, in a single file dedicated to the program in question, or in multiple coordinated files, e.g. B. Files that store on one or more modules, subprograms, or sections of code. A computer program may be employed to run on one computer or on multiple computers located at one location or distributed across multiple locations and linked by a data communications network.

The methods and logic flows described in this specification may be performed by one or more programmable computers executing one or more computer programs to perform functions by operating on input data and generating an output. The methods and logic flows can also be implemented by special purpose logic circuitry, e.g. FPGA or ASIC, or by a combination of special purpose logic circuitry and one or more programmed computers.

Suitable computers for executing a computer program may be based on general purpose or special purpose microprocessors or both, or on any other type of central processing unit. Generally, a central processing unit receives instructions and data from read-only memory or random access memory, or both. The essential elements of a computer are a central processing unit for executing instructions and one or more memory devices for storing instructions and data. The central processing unit and the memory can be supplemented by or integrated into special purpose logic circuitry. In general, a computer also includes one or more mass storage devices for storing data or is operatively connected to one or more mass storage devices for receiving data therefrom or transferring data thereto, e.g. B. magnetic, magneto-optical or optical disks. However, a computer need not have such devices. In addition, a computer in another device be embedded, e.g. a cellular phone, a personal digital assistant (PDA), a mobile audio or video player, a games console, a global positioning system (GPS) receiver, or a portable storage device, e.g. B. a universal serial bus (USB) stick, just to name a few.

Suitable computer-readable media for storing computer program instructions and data include all forms of non-transitory memory, media and storage devices, including, by way of example, semiconductor memory devices, e.g. B. EPROM, EEPROM and flash memory devices; magnetic discs, e.g. B. internal hard drives or removable drives; magneto-optical discs; and CD-ROM and DVD-ROM discs.

In order to provide interaction with a user, embodiments of the content described in this specification may be run on a computer having a display device, e.g. a CRT (cathode ray tube) or LCD (liquid crystal display) monitor, for displaying information to the user, and a keyboard and pointing device, e.g. a mouse or trackball, through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with a user; e.g. B. Feedback provided to the user can be any form of sensory feedback, e.g. B. visual feedback, audible feedback or tactile feedback; and input from the user can be received in any form, including auditory, verbal, or tactile input. Additionally, a computer may interact with a user by sending documents and receiving documents from a device used by the user; e.g. B. by sending web pages to a web browser on a user's device in response to requests received from the web browser. In addition, a computer can communicate with a user by sending text messages or other forms of messages to a personal device, e.g. a smartphone running a messaging application and receiving replying messages from the user.

Data processing devices, such as controller 100, for implementing machine learning models may also include, for example, special purpose hardware accelerators for processing common and computationally intensive parts of machine learning training or production, i. H. interference, loads, include.

Machine learning models can be executed and applied using a machine learning framework, e.g. For example, a TensorFlow framework, a Microsoft Cognitive Toolkit framework, an Apache Singa framework, or an Apache MXNet framework.

Embodiments of the content described in this specification may be executed on a computer system that includes a back-end component, e.g. as a data server, or comprising a middleware component, e.g. an application server, or that includes a front-end component, e.g. B. a client computer with a graphical user interface, web browser or application through which a user can interact with an execution of the content described in this description, or any combination of one or more such back-end, middleware or Front End Components. The components of the system may be connected by any form or medium of digital data communication, e.g. B. a communication network. Examples of communication networks include a local area network (LAN) and a wide area network (WAN), e.g. B. the Internet.

The computer system can include clients (the biomass heating system 1 ) and at least one server 190 . A client and server are generally remote from each other and typically interact through a communications network. The client and server relationship is created by computer programs running on the respective computers having a client-server relationship with each other.

The biomass heating system disclosed herein serves to burn biogenic solid fuel. However, the basic principles of the teaching explained herein (in particular relating to AI) can also be applied to gaseous fuel. For example, the present AI model (or the teaching of claim 1) can also be applied to a heating system that burns product gas (e.g. from a biomass gasifier). Here, for example, the methods disclosed herein and also the basic principles of learning the AI model 104 and the execution of the control can also be used.

1

Biomasse-Heizanlage

11

Kessel

12

Kesselfuß

13

Kesselgehäuse

14

Wasserzirkulationseinrichtung

15

Gebläse

16

Außenverkleidung

100

Steuereinrichtung der Biomasse-Heizanlage 1

111

Abgasstempertursensor

112

Lambdasonde

113

Unterdrucksensor oder Druckdifferenzsensor

114

Rücklauftemperatursensor bzw. Heizungswassertemperatursensor

115

Kesseltemperatursensor

116

Brennstoffbetthöhensensor

117

Brennkammertemperatursensor

101

maschinelle Lerneinheit

102

Userinterface bzw. Benutzerschnittstelle

104

KI-Modell

105, 105a, 105b

Speicher (lokal oder zentral)

106

Parameterdaten

107

Regelalgorithmus

190

Server

191

weitere (vorzugsweise externe) Datenquelle bzw. Sensoren

198

Verbindung

199

Netzwerkverbindung

2

Brenneinrichtung

21

erste Wartungsöffnung für die Brenneinrichtung

22

Drehmechanikhalterung

23

Drehmechanik

24

Brennkammer

25

Drehrost

26

Primärverbrennungszone der Brennkammer

27

Sekundärverbrennungszone bzw. Strahlungsteil der Brennkammer

28

Brennstoffbett

29

Brennkammersteine

A1

erste Horizontalschnittlinie

A2

erste Vertikalschnittlinie

201

Zündeinrichtung

202

Brennkammerschräge

203

Brennkammerdüse

204

Brennkammerdecke

211

Dämmmaterial bspw. Vermiculite

231

Antrieb bzw. Motor(en) der Drehmechanik

251

Bodenplatte des Drehrosts

252

Erstes Drehrostelement

253

Zweites Drehrostelement

254

Drittes Drehrostelement

255

Übergangselement

256

Öffnungen

257

Rostlippen

258

Verbrennungsfläche

260

Auflageflächen der Brennkammersteine

261

Nut

262

Vorsprung

263

Ring

264

Halterungssteine

265

Schräge der Halterungssteine

291

Sekundärluft bzw. Rezirkulationsdüsen

3

Wärmetauscher

31

Wartungsöffnung für Wärmetauscher

32

Kesselrohre

33

Kesselrohreintritt

34

Wendekammereintritt

35

Wendekammer

36

Federturbulator

37

Band- oder Spiralturbulator

38

Wärmetauschmedium

331

Isolation am Kesselrohreintritt

4

Filtereinrichtung

41

Abgasausgang

42

Elektrodenversorgungsleitung

43

Elektrodenhalterung

44

Filtereintritt

45

Elektrode

46

Elektrodenisolation

47

Filteraustritt

48

Käfig

49

Rauchgaskondensator

411

Rauchgaszuleitung zum Rauchgaskondensator

412

Rauchgasausgang aus dem Rauchgaskondensator

481

Käfighalterung

491

erster Fluidanschluss

491

zweiter Fluidanschluss

493

Wärmetauscherrohr

4931

Rohrhalteelement

4932

Rohrbodenelement

4933

Schlaufen/Umkehrstellen

4934

erste Zwischenräume der Wärmetauscherrohre zueinander

4935

zweite Zwischenräume der Wärmetauscherrohre zu der Außenwand des Rauchgaskondensators

4936

Durchlässe

495

Kopfelement

4951

Kopfelementströmungsführung

496

Kondensataustritt

4961

Kondensatsammeltrichter

497

Flansch

498

Seitenfläche mit Wartungsöffnung

499

Halterungseinrichtung für den Rauchgaskondensator

5

Rezirkulationseinrichtung

50

Ringkanal um Brennkammersteine

52

Luftventil

52s

Schieberventil

53

Rezirkulationseintritt

54

Primärmischkanal

55

Sekundärmischkanal

55a

Sekundärtemperierungskanal

56

Primärrezirkulationskanal

57

Sekundärrezirkulationskanal

58

Primärluftkanal

59

Sekundärluftkanal

5a

Primärmischeinheit

5b

Sekundärmischeinheit

521

Ventilstellaktor

522

Ventilstellachsen

523

Ventilflügel

524

Ventilgehäuse

525

Ventilvorkammer

526

Ventildurchtrittsöffnung

527

Ventilkörper

528

Ventilfläche

531

Rezirkulationseintrittskanal

532

Rezirkulationseintrittskanalteiler

541

Primärdurchtritt

542

Primärmischkammer

543

Primärmischkammeraustritt

544

Primärreziventileintritt

545

Primärluftventileintritt

546

Primärmischkammergehäuse

551

Sekundärdurchtritt

552

Sekundärmischkammer

553

Sekundärmischkammeraustritt

554

Sekundärreziventileintritt

555

Sekundärluftventileintritt

556

Sekundärmischkammergehäuse

581

Primärlufteintritt

582

Primärluftsensor

591

Sekundärlufteintritt

592

Sekundärluftsensor

6

Brennstoffzufuhr

61

Zellradschleuse

62

Achse der Brennstoffzufuhr

63

Übersetzungsmechanik

64

Brennstoffzufuhrkanal

65

Brennstoffzufuhröffnung

66

Antriebsmotor

67

Brennstoff-Förderschnecke

7

Ascheabfuhr

71

Ascheaustragungsschnecke

711

Schneckenachse

712

Zentrierungsscheibe

713

Wärmetauscherabschnitt

714

Brennerabschnitt

72

Motor der Ascheabfuhr mit Mechanik

73

Übergangsschnecke

731

rechter Unterabschnitt - nach links steigende Schnecke

732

linker Unterabschnitt -nach rechts steigende Schnecke

74

Aschebehälter

75

Übergangschneckengehäuse

751

Öffnung des Übergangsschneckengehäuses

752

Begrenzungsblech

753

Hauptkörperabschnitt des Gehäuses

754

Befestigungs- und Trennelement

755

Trichterelement

81

Lagerachsen

82

Drehachse der Brennstoff-Niveauklappe

83

Brennstoff-Niveauklappe

831

Hauptfläche

832

Mittenachse

833

Oberflächenparallele

834

Öffnungen

84

Lagerkerbe

85

Sensorflansch

86

Glutbetthöhenmessmechanik

9

Reinigungseinrichtung

91

Reinigungsantrieb

92

Reinigungswellen

93

Wellenhalterung

94

Fortsatz

95

Turbulatorhalterungen

951

Drehlageraufnahme

952

Fortsätze

953

Durchlässe

954

Ausnehmungen

955

Drehlagergestänge

96

zweiarmiger Schlaghebel

97

Anschlagkopf

E

Einschubrichtung des Brennstoffs

S∗

Strömungspfeile

The embodiments and aspects disclosed herein were provided for description and understanding of the disclosed technical matters and are not intended to limit the scope of the present disclosure. Therefore, it is to be construed that the scope of the present disclosure includes any modification or other various embodiments based on the technical spirit of the present disclosure.

1

Biomass heating system

11

boiler

12

boiler foot

13

boiler housing

14

water circulation device

15

fan

16

outer lining

100

Control device of the biomass heating system 1

111

exhaust gas temperature sensor

112

lambda probe

113

Vacuum sensor or pressure difference sensor

114

Return temperature sensor or heating water temperature sensor

115

boiler temperature sensor

116

fuel bed height sensor

117

combustion chamber temperature sensor

101

machine learning unit

102

User interface or user interface

104

AI model

105, 105a, 105b

Storage (local or central)

106

parameter data

107

control algorithm

190

server

191

further (preferably external) data source or sensors

198

connection

199

network connection

2

burner

21

first maintenance opening for the burner

22

rotary mechanism mount

23

turning mechanism

24

combustion chamber

25

rotary grate

26

Primary combustion zone of the combustor

27

Secondary combustion zone or radiant part of the combustion chamber

28

fuel bed

29

combustion chamber bricks

A1

first horizontal cutting line

A2

first vertical cutting line

201

ignition device

202

combustion chamber slope

203

combustor nozzle

204

combustor ceiling

211

Insulation material e.g. vermiculite

231

Drive or motor(s) of the turning mechanism

251

bottom plate of the rotary grate

252

First rotary grate element

253

Second rotary grate element

254

Third rotary grate element

255

transition element

256

openings

257

rust lips

258

burn surface

260

Support surfaces of the combustion chamber bricks

261

groove

262

head Start

263

ring

264

bracket stones

265

slope of the support stones

291

Secondary air or recirculation nozzles

3

heat exchanger

31

Maintenance opening for heat exchanger

32

boiler tubes

33

Boiler tube entry

34

reversing chamber entry

35

turning chamber

36

spring turbulator

37

Band or spiral turbulator

38

heat exchange medium

331

Insulation at the boiler tube inlet

4

filter device

41

exhaust outlet

42

electrode supply line

43

electrode holder

44

filter inlet

45

electrode

46

electrode insulation

47

filter outlet

48

Cage

49

flue gas condenser

411

Flue gas supply line to the flue gas condenser

412

Flue gas outlet from the flue gas condenser

481

cage mount

491

first fluid connection

491

second fluid port

493

heat exchanger tube

4931

pipe holding element

4932

tube sheet element

4933

loops/reversals

4934

first spaces between the heat exchanger tubes

4935

second spaces of the heat exchanger tubes to the outer wall of the flue gas condenser

4936

culverts

495

header element

4951

header flow guide

496

condensate outlet

4961

condensate collection funnel

497

flange

498

Side surface with maintenance opening

499

Mounting device for the flue gas condenser

5

recirculation device

50

Ring channel around combustion chamber bricks

52

air valve

52s

slide valve

53

recirculation entry

54

primary mixing channel

55

secondary mixing channel

55a

secondary tempering channel

56

primary recirculation channel

57

secondary recirculation channel

58

primary air duct

59

secondary air duct

5a

primary mixing unit

5b

secondary mixing unit

521

valve actuator

522

valve positioning axes

523

valve vane

524

valve body

525

valve antechamber

526

valve passage opening

527

valve body

528

valve face

531

recirculation entry channel

532

recirculation entry channel splitter

541

primary passage

542

primary mixing chamber

543

primary mixing chamber exit

544

primary reciprocating valve entry

545

Primary air valve inlet

546

Primary mixing chamber housing

551

secondary passage

552

secondary mixing chamber

553

secondary mixing chamber outlet

554

secondary reciprocating valve entry

555

secondary air valve inlet

556

secondary mixing chamber housing

581

primary air inlet

582

primary air sensor

591

secondary air intake

592

secondary air sensor

6

fuel supply

61

rotary valve

62

Fuel supply axis

63

translation mechanics

64

fuel supply channel

65

fuel supply opening

66

drive motor

67

fuel auger

7

ash removal

71

ash discharge screw

711

worm axis

712

centering disc

713

heat exchanger section

714

burner section

72

Ash removal motor with mechanics

73

transition snail

731

right subsection - snail rising to the left

732

left subsection - snail rising to the right

74

ash container

75

Transition snail shell

751

Opening of the transition worm housing

752

boundary plate

753

Main body section of the housing

754

Fastening and separating element

755

funnel element

81

bearing axles

82

Axis of rotation of the fuel level flap

83

fuel level flap

831

main surface

832

center axis

833

surface parallel

834

openings

84

bearing notch

85

sensor flange

86

Ember bed height measurement mechanism

9

cleaning device

91

cleaning drive

92

cleaning shafts

93

shaft mount

94

extension

95

turbulator mounts

951

Pivot bearing mount

952

appendages

953

culverts

954

recesses

955

pivot linkage

96

two-armed hammer

97

stop head

E

insertion direction of the fuel

S∗

flow arrows

Claims (22)

Biomass heating system (1) for firing biogenic fuel, having: a boiler (11) with a burner (2) and with a heat exchanger (3); a controller (100) having a memory (105, 105a); at least one sensor (86, 111-117, 582, 592) for providing sensor data, which can detect at least one chemical and/or physical variable of the biomass heating system (1) and which is communicatively connected to the control device (100); at least one actuator (4, 5, 52, 6, 61, 66, 7, 72, 91, 201, 231) of the biomass heating system (1), which is communicatively connected to the control device (100) and can be controlled by it ; wherein the control device (100) contains at least one AI model (104) for controlling the biomass heating system (1) based on the sensor data provided, the AI model (104) being parameterized using machine learning. Biomass heating system (1) according to claim 1, wherein
the AI model is a neural network with at least 3 layers and at least 200 neurons. Biomass heating system (1) according to claim 1 or 2, wherein the control device (100) is set up in such a way that the sensor data are aggregated over a predetermined period of time during operation of the biomass heating system (1), and the aggregated sensor data is used for machine learning. Biomass heating system (1) according to claim 1 or 2 or 3, wherein the control device (100) of the biomass heating system (1) has a machine learning unit (101); and the biomass heating system (1) is set up in such a way that the variable detected by the at least one sensor (86, 111-117, 582, 592) is stored as the sensor data in the memory (105a) of the control device (100); and the machine learning takes place by means of the machine learning unit (101) using the sensor data stored in the memory (105a). Biomass heating system (1) according to claim 1 or 2 or 3, wherein
the biomass heating system (1) is set up in such a way that the at least one variable detected by the at least one sensor (86, 111-117, 582, 592) is stored as the sensor data in the memory (105a) of the control device (100); and the sensor data stored in the memory (105a) of the control device (100) are transmitted to a central processing unit (190) via a network connection (199); and an AI model (104) or weight data for controlling the biomass heating system (1) is/are transmitted from the central processing unit (190) to the control device (100) of the biomass heating system (1), the AI model (104 ) is parameterized by machine learning by means of a machine learning unit of the central processing unit (190) at least using the transmitted sensor data. Biomass heating system (1) according to any one of the preceding claims, wherein the AI model (104) is set up to classify the fuel, and the classification of the fuel is based on the following sensor data:
an oxygen content of the exhaust gas of the biomass heating system (1), which is detected with a lambda probe (112); and at least one of the following: an exhaust gas temperature of the exhaust gas of the biomass heating system (1), which is detected with an exhaust gas temperature sensor (111); a combustion chamber temperature of a combustion chamber (24) of the biomass heating system (1), which is detected with a combustion chamber temperature sensor (117); wherein the oxygen content, the exhaust gas temperature and the combustion chamber temperature are recorded in a period of time after ignition of the fuel by an ignition device (201) until a predefined oxygen content, preferably 16% by volume O ₂ , is reached. Biomass heating system (1) according to any one of the preceding claims, wherein
the control device (100) is set up with the AI model (104) to regulate a motor (66) of a fuel feed (6) into the combustion device (2) to a specified fuel bed height, the regulation being based on at least one of the following sensor data :
an oxygen content of the exhaust gas of the biomass heating system (1), which is detected with a lambda probe (112); and at least one of the following: an exhaust gas temperature of the exhaust gas of the biomass heating system (1), which is detected with an exhaust gas temperature sensor (111); a combustion chamber temperature of a combustion chamber (24) of the biomass heating system (1), which is detected with a combustion chamber temperature sensor (117). Biomass heating system (1) according to any one of the preceding claims, wherein
the control device (100) is set up with the AI model (104) to carry out lambda control of the biomass heating system to a predetermined lambda value of the exhaust gas, the control being based on at least one of the following sensor data:
an oxygen content of the exhaust gas of the biomass heating system (1), which is detected with a lambda probe (112); and at least one of the following: an exhaust gas temperature of the exhaust gas of the biomass heating system (1), which is detected with an exhaust gas temperature sensor (111); a combustion chamber temperature of a combustion chamber (24) of the biomass heating system (1), which is detected with a combustion chamber temperature sensor (117); a negative pressure in the combustion chamber (24) which is detected by a negative pressure sensor (113); a fuel bed level on the rotary grate (25) sensed by a fuel bed level sensor (116); and wherein at least one of the air valves (52) of the recirculation device (5) is controlled by the AI model (104). Biomass heating system (1) according to any one of the preceding claims, wherein
the control device (100) is set up with the AI model (104) to regulate the output of the biomass heating system to a predetermined output value of the biomass heating system (1), the control being based on at least one of the following sensor data and/or external Data based: a combustion chamber temperature of a combustion chamber (24) of the biomass heating system (1), which is detected with a combustion chamber temperature sensor (117); a fuel bed level on the rotary grate (25) sensed by a fuel bed level sensor (116); a boiler temperature which is a temperature of the heat exchange medium (38) in the boiler (11); an amount of heat given off by the boiler (11), which is detected by means of an amount of heat sensor; Weather forecast data retrieved over a network (199), preferably the Internet; and and wherein at least one of the air valves (52) of the recirculation device (5) and/or the primary air supply is controlled by the AI model (104), or the blower (11) of the biomass heating system is controlled. Biomass heating system (1) according to any one of the preceding claims, wherein
the control device (100) is set up with the AI model (104) to regulate the position of a valve (52) of a primary air duct (58) of the biomass heating system (1) to a predetermined amount of primary air, the regulation being based on at least one of the based on the following sensor data: a combustion chamber temperature of a combustion chamber (24) of the biomass heating system (1), which is detected with a combustion chamber temperature sensor (117); a fuel bed level on the rotary grate (25) sensed by a fuel bed level sensor (116); an oxygen content of the exhaust gas of the biomass heating system (1), which is detected with a lambda probe (112); a boiler temperature which is a temperature of the heat exchange medium (38) in the boiler (11); an amount of heat given off by the boiler (11), which is detected by means of an amount of heat sensor; Biomass heating system (1) according to any one of the preceding claims, wherein the control device (100) is set up with the AI model (104) to carry out on/off control of the biomass heating system, the control being based on the following: Weather forecast data retrieved over a network (199), preferably the Internet; and a predetermined boiler setpoint temperature; a boiler temperature which is a temperature of the heat exchange medium (38) in the boiler (11); and optionally based on at least one of the following sensor data: a temperature of a heat accumulator, preferably a buffer, in a heating circuit which is supplied with energy by the biomass heating system (1); a power output capability of the biomass heating system (1), for example the power class of the boiler (kW); an amount of heat given off by the boiler (11), which is detected by means of an amount of heat sensor. Biomass heating system (1) according to one of the preceding claims,
wherein the result of the calculation of the trained AI model (104) is used as at least one input variable for a control algorithm (107), for example a P, PI, PID or PD control. Method for controlling a biomass heating system for firing solid fuel in the form of pellets and/or wood chips, the biomass heating system (1) having: a boiler (11) with a burner (2) and a heat exchanger (3); a controller (100) having a memory (105, 105a); the method comprising:
Controlling the biomass heating system (1) using an AI model (104) of the control device (100), which is parameterized by machine learning. A method according to claim 13, wherein
controlling the biomass heating system (1) includes: Detection of at least one chemical and / or physical variable of the biomass heating system (1) by at least one sensor (111-117, 582, 592) of Biomass heating system (1), which is communicatively connected to the control device (100); and processing the at least one chemical and/or physical variable using the AI model; and Controlling the at least one actuator (4, 5, 52, 6, 61, 66, 7, 72, 91, 201, 231) of the biomass heating system (1), which is communicatively connected to the control device (100), based on the Result of the processing step. The method of claim 13 or 14, the method further comprising the steps of: Aggregating the sensor data over a predetermined period of time during operation of the biomass heating system (1), and Using the aggregated sensor data for machine learning. A method according to claim 13, 14 or 15, wherein the control device (100) of the biomass heating system (1) has a machine learning unit (101); and the method further comprises the following step:
storing the at least one variable detected by the plurality of sensors (86, 111-117, 582, 592) in the memory (105a) of the control device (100) as the sensor data; wherein the machine learning takes place by means of the machine learning unit (101) using the sensor data stored in the memory (105). A method according to any one of claims 13 to 16, the method further comprising the steps of: storing the at least one variable detected by the at least one sensor (86, 111-117, 582, 592) in the memory (105a) of the control device (100) as the sensor data; and Transmission of the sensor data stored in the memory (105a) of the control device (100) via a network connection (199) to a central processing unit (190); and Parameterizing the AI model (104) by machine learning using a machine learning unit of the central processing unit (190) using the transmitted sensor data; and Transmission of the parameterized AI model (104) to the control device (100) of the biomass heating system (1) from the central processing unit (190). Method for controlling a biomass heating system (1) for firing solid fuel in the form of pellets and/or wood chips, having the biomass heating system (1) and the processes according to one of Claims 1 to 12. Method for a system with a plurality of biomass heating systems (1) and with at least one central computing device (190), wherein the biomass heating systems (1) are communicatively connected to the central computing device (190) via a network (199), wherein
the majority of biomass heating systems (1) have the following: a boiler (11) with a burner (2) and a heat exchanger (3); a controller (100) having a memory (105, 105a); a plurality of sensors (86, 111-117, 582, 592) for providing sensor data which can detect chemical and/or physical parameters of the biomass heating system (1) and which are communicatively connected to the control device (100); a plurality of actuators (4, 5, 52, 6, 61, 66, 7, 72, 91, 201, 231) of the biomass heating system (1), which are communicatively connected to the control device (100) and are controlled by it be able; the method comprising the following steps: Detection of at least one chemical and/or physical variable of the biomass heating system (1) by at least one sensor (86, 111-117, 582), 592 of the biomass heating system (1); and Storing the detected values in the memories (105a) of the control devices (100) of the biomass heating systems (1) as sensor data; Transmission of at least some of the variables stored in the memories (105a) of the control devices (100) of the biomass heating systems (1) via at least one network connection (199) to the central processing unit (190). The method according to claim 19, further comprising the following steps: Aggregating the transmitted sensor data in the central processing unit; Machine learning by means of the machine learning unit (101) using the aggregated sensor data, resulting in a parameterized AI model (104); Transmission of the parameterized AI model (104) to the control device (100) of the biomass heating system (1) from the central processing unit (190) via the network (199); Controlling the biomass heating system (1) by means of the transmitted AI model (104) by the control device (100). A method according to claim 19 or 20, further comprising the features of at least one of claims 1 to 3 and 6 to 12. A computer-readable storage medium containing instructions which, when executed by a computing device, cause the computing device to perform any of the methods of claims 13 to 21.

Images