KR102381773B1 - Training data generating method for training health monitering model of power plant boiler and device therefor - Google Patents

Abstract

The present invention relates to a method and apparatus for generating training data for training a health monitoring model of a boiler in a power plant, the method comprising: power plant operation data including boiler condition measurement data, fuel property data, and power plant output data from a power plant Separating the output and fuel maintenance section combining the obtaining step, the output maintenance section in which the power plant output data is kept constant, and the fuel maintenance section in which the type of fuel is kept constant among the combustion sections of the fuel included in the fuel property data Step, in the output and fuel maintenance section, determining the boiler health index according to the flow rate formation time based on the flow data of the reheater overheating reduction water included in the boiler condition measurement data, and the operation of the boiler among the power plant operation data For a preset number of boiler operation variables selected in advance to determine the state, based on the boiler health index, generating training data including representative values of the boiler operation variables calculated for each preset cycle, The data may include a representative value of a boiler operation variable for each preset cycle as an input value, and may include a boiler health index in the corresponding cycle determined in the step of determining the boiler health index as a label.

Description

TRAINING DATA GENERATING METHOD FOR TRAINING HEALTH MONITERING MODEL OF POWER PLANT BOILER AND DEVICE THEREFOR

The present invention is a method of generating training data for training a plant/system health monitoring model for a power plant boiler based on artificial intelligence so as to minimize dependence on the failure mechanism and physical characteristics of the plant/system for the power plant boiler. and devices.

The research related to the present invention is supported by the Korea Electric Power Corporation's 2018 selection of basic research and development project research funds (task name: "Development of analysis technology for monitoring and diagnosis support of facility/system health based on power plant big data for the realization of smart power plants") Applied (task number: R18XA06-46).

In general, a method for determining whether a power plant boiler system is abnormal can be roughly divided into two types.

First, when a trip logic condition according to a trip diagram input to the control system in advance is satisfied, a response is made by stopping a specific facility or the entire power plant. However, if the trip logic condition is satisfied, the facility or power plant has already been shut down, which is a follow-up measure, so loss is unavoidable.

Second, if an abnormal situation is detected through the alarm of the boiler monitoring and monitoring system by a professional operator, the problem is recovered through appropriate operation control. However, in the case of monitoring by a professional operator, there is a deviation in monitoring capability based on experience, and there is a disadvantage in that the monitoring range is limited because it depends on people. In the case of a monitoring system, an alarm is set to sound when a predefined set point or normal section is exceeded. There are limitations in judging whether or not there is an abnormality.

The above-mentioned background art is technical information that the inventor possessed for the derivation of the present invention or acquired in the process of derivation of the present invention, and cannot necessarily be said to be a known technique disclosed to the general public prior to the filing of the present invention.

One object of the present invention is to solve the problem of the prior art, which is a post-action measure in which a loss is inevitable by stopping a specific facility or the entire power plant according to a trip logic condition pre-entered into the control system as a method for determining whether a power plant boiler system is abnormal. it is to do

One object of the present invention is to detect an abnormal situation through boiler monitoring by a professional operator as a method for determining whether a boiler system is abnormal in a power plant, and in the case of monitoring by a professional operator, there is a deviation in the monitoring capability based on experience and the monitoring range is limited This is to solve the problems of the prior art.

An object of the present invention is to provide a method for determining whether a power plant boiler system is abnormal. When an abnormal situation is detected through an alarm of the monitoring system, false alarms are frequent, so there is a limitation in monitoring the overall health of the facility and determining whether there is an abnormality. It is to solve technical problems.

An object of the present invention is to train a boiler facility/system health monitoring model based on training data that minimizes dependence on failure mechanisms and physical characteristics of facilities/systems based on power plant big data.

An object of the present invention is to detect abnormalities and changes in a boiler and support diagnosis (identification of the cause) using a boiler facility/system health monitoring model based on artificial intelligence.

The problem to be solved by the present invention is not limited to the above-mentioned problems, and other problems and advantages of the present invention not mentioned can be understood by the following description, and more clearly understood by the embodiments of the present invention will be In addition, it will be understood that the problems and advantages to be solved by the present invention can be realized by means and combinations thereof indicated in the claims.

A method for generating training data for training a health monitoring model of a boiler in a power plant according to an embodiment of the present invention includes boiler state measurement data measured by a sensor from a power plant, fuel properties data, and power plant operation including power plant output data An output and fuel maintenance section combining the steps of acquiring data, an output maintenance section in which the power plant output data is kept constant, and a fuel maintenance section in which the type of fuel is kept constant among the combustion sections of fuel included in the fuel property data The step of classifying, the step of determining the boiler health index according to the flow rate formation time based on the flow data of the reheater overheating reduction water included in the boiler status measurement data in the output and fuel maintenance section, and the boiler among the power plant operation data For a preset number of boiler operation variables selected in advance to determine the operating state of , the training data may include a representative value of a boiler operation variable for each preset cycle as an input value, and may include a boiler health index in the corresponding cycle determined in the step of determining the boiler health index as a label.

A training data generating apparatus for training a health monitoring model of a power plant boiler according to an embodiment of the present invention includes boiler state measurement data measured by a sensor from a power plant, fuel properties data, and power plant operation including power plant output data An output and fuel maintenance section combining an acquisition unit for acquiring data, an output maintenance section in which the power plant output data is kept constant, and a fuel maintenance section in which the type of fuel is kept constant among the combustion sections of fuel included in the fuel property data A data mart component that separates For a preset number of boiler operation variables pre-selected to determine the operation state of the boiler among the power plant operation data, based on the boiler health index, training data including representative values of the boiler operation variables calculated for each preset cycle is generated. The training data includes a representative value of the boiler operation variable for each preset cycle as an input value, and the boiler health index in the corresponding cycle determined in the step of determining the boiler health index can be included as a label. .

In addition to this, another method for implementing the present invention, another system, and a computer-readable recording medium storing a computer program for executing the method may be further provided.

Other aspects, features and advantages other than those described above will become apparent from the following drawings, claims, and detailed description of the invention.

According to the present invention, based on the power plant big data, training data for training the health monitoring model of the power plant boiler in an environment that minimizes dependence on the physical characteristics of facilities/systems and the failure mechanism is generated in the actual power plant operating environment. Boiler health monitoring performance for data can be improved.

In addition, it is possible to increase the utilization of dark data (data that only collects/stores and does not use), which has not been utilized so far, to improve the operator's operating capability and expand the scope of monitoring.

In addition, it can provide information to quickly search for areas that require precise diagnosis and action, thereby minimizing unplanned detection and trips, thereby contributing to improved availability and efficiency.

Effects of the present invention are not limited to those mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the following description.

1 is a table comparing the power plant monitoring method according to the present embodiment.
FIG. 2 is a table comparing two approaches of the data-based monitoring method of FIG. 1 .
3 is an exemplary diagram of a health monitoring environment including a power plant, a boiler health monitoring apparatus, and a network connecting them according to the present embodiment.
4 is a schematic diagram of a power plant and a cross-sectional view of a boiler in FIG.
FIG. 5 is a diagram schematically illustrating a configuration of a boiler health monitoring device including a training data generating device for training a power plant boiler health monitoring model in FIG. 3 .
6 is a diagram schematically illustrating data obtained by the training data generating apparatus of FIG. 5 from a power plant.
7 is a diagram schematically illustrating an output and fuel maintenance section extracted from data collected by the training data generating apparatus of FIG. 5 .
FIG. 8 is a diagram schematically illustrating the criteria for determining the health of the boiler determined by the training data generating apparatus of FIG. 5 .
9 is a diagram schematically illustrating a boiler operation variable selected by the training data generating apparatus in FIG. 5 to generate training data.
10 is a diagram schematically illustrating a final analysis mart determined by the training data generating apparatus in FIG. 5 .
11 is a diagram schematically illustrating health detection and boiler health monitoring model learning and verification performed by the boiler health monitoring device in FIG. 3 .
12 is a flowchart for explaining a training data generation method for training a health monitoring model of a power plant boiler according to the present embodiment.

Advantages and features of the present invention, and a method for achieving them will become apparent with reference to the detailed description in conjunction with the accompanying drawings. However, the present invention is not limited to the embodiments presented below, it can be implemented in a variety of different forms, and should be understood to include all transformations, equivalents, and substitutes included in the spirit and scope of the present invention. . The embodiments presented below are provided so that the disclosure of the present invention is complete, and to completely inform those of ordinary skill in the art to which the present invention pertains to the scope of the invention. In describing the present invention, if it is determined that a detailed description of a related known technology may obscure the gist of the present invention, the detailed description thereof will be omitted.

The terms used in the present application are only used to describe specific embodiments, and are not intended to limit the present invention. The singular expression includes the plural expression unless the context clearly dictates otherwise. In the present application, terms such as “comprise” or “have” are intended to designate that a feature, number, step, operation, component, part, or combination thereof described in the specification exists, but one or more other features It should be understood that this does not preclude the existence or addition of numbers, steps, operations, components, parts, or combinations thereof. Terms such as first, second, etc. may be used to describe various elements, but the elements should not be limited by the terms. The above terms are used only for the purpose of distinguishing one component from another.

Hereinafter, embodiments according to the present invention will be described in detail with reference to the accompanying drawings. decide to do

1 is a table comparing the monitoring method of a power plant according to the present embodiment, and FIG. 2 is a table comparing two approaches of the data-based monitoring method in FIG. 1 .

Domestic power generation companies are actively building smart power plants using core technologies of the 4th industrial revolution (IoT, big data, AI, etc.). A smart power plant aims to realize an intelligent power plant that analyzes data and operation information accumulated from numerous sensors and makes an optimized decision by judging the operating status of the power plant in real time.

Overseas advanced companies are entering the digital solution-based operation improvement and service business by securing data analysis source technology based on their main equipment design/manufacturing capabilities, whereas the domestic power generation industry has relatively low investment in big data analysis source technology. It is insufficient.

Sensors and operation/maintenance data (hereafter, power plant big data) accumulated over a long period of time in power plants contain information on unique characteristics and state changes, so efficient use and analysis of power plant big data is an essential task for realizing smart power plants.

The basis of implementing a smart power plant is to monitor the health of objects of interest (facilities, systems, etc.) and to provide information that can take action by detecting or predicting abnormalities/changes. This can be the basis for implementing optimal predictive maintenance and operational optimization.

Many system monitoring and diagnosis technologies currently developed correspond to the physics-driven approach of FIG. 1 . This approach mainly targets specific areas/parts or failure modes for monitoring, and often requires high sensor resolution and specialized analysis techniques. On the other hand, power plant big data is collected for the purpose of grasping or controlling the system status, except for some major facilities, and is stored by adjusting the resolution (data measurement period, accuracy or precision level) for long-term retention. Power plant big data reduces the capacity burden by recording a value when a change above the standard value occurs or by storing the measurement cycle in minutes/second units.

In addition, since there are no sensors in all parts, in the specialized/segmented market, it is difficult to apply big data to power plants for monitoring and diagnosis technology, or additional facility expansion is required for implementation, and large operating costs are often required. For example, high-resolution vibration measurement/analysis facilities are separately installed for turbine/generator vibration analysis, but long-term data storage is not possible due to capacity problems, and analysis takes a lot of time and money.

For this reason, the data-driven approach is evaluated to be advantageous to efficiently utilize the power plant big data, and this approach mainly adopts the facility/system unit as the target of the monitoring scope. At this time, when monitoring a facility/system unit using a data-based approach, the approach is divided into two types as shown in FIG. 2 .

In FIG. 2 , the residual prediction method is widely adopted in the conventional anomaly detection and diagnosis methodology to comprehensively monitor facilities/systems. However, due to the approach of estimating the value of each sensor independently, there is a limit to comprehensively analyzing the equipment/system of a power plant that moves with the close relationship between the preceding and following processes and the complex behavior.

Therefore, there is a need for a method that analyzes the overall health of facilities/systems using big data of power plants, detects abnormalities/changes based on this, and acts as a compass for cause diagnosis. In other words, it is necessary to supplement the existing monitoring/diagnostic technology with a top-down method of data analysis that looks at the forest and searches for problem areas.

This embodiment develops a data-based facility/system health monitoring model that minimizes dependence on the physical characteristics of facilities/systems and failure mechanisms based on the power plant big data accumulated so far, and detects and diagnoses abnormalities and changes (causes to support understanding).

Through this, it is possible to increase the utilization of power plant dark data that has not been utilized so far (data that only collects/stores and does not use), thereby improving the operator's operating capability and expanding the monitoring range. In addition, it can provide information to quickly search for areas that require precise diagnosis and action, thereby contributing to the improvement of availability and efficiency by minimizing unplanned detection and trips.

For convenience of description, the subject of the present embodiment may be set as a boiler of a coal-fired power plant among power plants using fuel. Among boilers, turbines, and generators, which are the main equipment of a power plant, the boiler has a non-rotating body, and energy that is connected to fuel (coal), air (air), combustion gas (combustion/flue gas), and feedwater/steam It is a device that generates high-temperature/high-pressure steam that rotates a turbine through conversion.

The anomaly detection and diagnosis method of the rotating body (turbine/generator) main machine can be analyzed based on relatively well-established failure mechanisms and theories, and the physics failure-based approach is advantageous. On the other hand, it is known that the cause of anomalies is diverse and it is difficult to predict the actual operating environment because boilers have high uncertainty in analysis due to very complex physical/chemical reactions and incidental phenomena that occur during the combustion process and operate by combining various systems. there is.

Therefore, rather than relying entirely on a physical failure-based approach for detecting and diagnosing boiler abnormalities, it is a fusion of data science and artificial intelligence technology with domain knowledge of boiler design and operation characteristics. method is efficient.

In particular, sensor and operation data accumulated over a long period of time in power plants (hereafter, power plant big data) contain information about the unique characteristics and state changes of boiler facilities and systems, but are not fully utilized and remain as dark data. Therefore, through this embodiment, the power plant big data is systematically utilized to monitor the status of the boiler and to diagnose the abnormality and cause.

3 is an exemplary diagram of a health monitoring environment including a power plant, a boiler health monitoring device, and a network connecting them according to the present embodiment, and FIG. 4 is a schematic diagram of the power plant and a cross-sectional view of the boiler in FIG. In the following description, descriptions of parts overlapping with those of FIGS. 1 and 2 will be omitted. 3 and 4 , the health monitoring environment may include a coal-fired power plant 100 , a boiler health monitoring device 200 , and a network 300 .

Coal-fired power plant 100 may generate electricity by burning fuel as fuel in a boiler to generate steam, and rotating a turbine using the generated steam. Hereinafter, the fuel will be described with reference to the fuel, but it will be clear that the contents described in the present disclosure may be equally or similarly applied to fuels other than the fuel.

The boiler system of the coal-fired power plant 100 can be largely divided into a fuel/air system, a combustion gas system, and a water supply/steam system according to an energy conversion stage. One of the key management indicators in boiler operation is to maintain constant high temperature/high pressure steam quality. All components of the boiler system are organically controlled to achieve this, and it is very important to stably maintain the temperature balance of feedwater/steam under various loads and fuel conditions.

Looking at the feedwater/steam system in the boiler system, the feedwater heated by the economizer passes through the furnace waterwall and absorbs energy from the radiant heat of the combustion gas. It can be converted into supercritical steam. Pure steam separated from water through a separator can be passed through a superheater to further increase the temperature and pressure of the steam to maximize thermal efficiency before being delivered to a high-pressure turbine. The steam that primarily rotates the turbine is recovered again and delivered to a reheater, where it is reheated and passes through an intermediate-pressure turbine and a low-pressure turbine, where it is secondarily rotated for fuel. of chemical energy can be finally converted into kinetic energy.

At this time, if the temperature of the steam rises above the set value, the temperature of the steam is lowered by operating the attemperator because it may adversely affect the life of the equipment such as a decrease in strength of the turbine or boiler tube, deterioration of high temperature, cracks, etc. can be done

The superheater is a facility that sprays water inside the tube to lower the temperature of steam, and is connected to a superheater and a reheater. The superheat reduction water supplied to the superheater does not affect the efficiency of the boiler regardless of the flow rate, but the overheat reduction water of the reheater may cause a loss in the efficiency of the boiler as the flow rate increases.

The direct cause of triggering the operation of the reheater overheater may include that the outlet steam temperature of the reheater rises above a certain level. The root cause of this phenomenon is very diverse, such as a change in fuel properties, a change in the amount of water/coal supply/air, combustion atmosphere, and slagging/fouling of the heat transfer part, and these causes can act in a complex way. .

In this case, a comprehensive index capable of determining soundness by using the big data of the boiler in the coal-fired power plant 100 may include a heat balance of the water supply/steam system. Since the temperature of the inlet/outlet for each part of the system has a design value, it represents the overall soundness based on the boiler operation variables (ex. output, coal feed, air, etc.), tube/steam temperature, and heat transfer surface temperature change. Feature information can be extracted.

For example, Economizer - Furnace Waterwall - Superheater - (High Pressure Turbine) - Reheater - (Medium Pressure Turbine) Preceding process of reheater in feedwater/steam system flow If the heat exchange of the combustion gas is not performed properly at The overall thermal balance may deviate from the steady state.

Therefore, in this embodiment, the overheating reduction flow rate of the reheater is formed due to the heat absorption imbalance of the feedwater/steam system, which causes the loss of boiler efficiency as abnormal, and based on the boiler design and operation characteristics As such, we would like to present a specific soundness monitoring methodology using power plant big data.

The boiler health monitoring apparatus 200 may acquire power plant operation data including boiler state measurement data measured by a sensor from the coal-fired power plant 100 , fuel properties data, and power plant output data.

Here, the boiler condition measurement data is the differential pressure, coal feed amount, BTU (British thermal unit), primary air volume, secondary air volume, burner angle, tube temperature for each part of the water/steam system, inlet/outlet steam temperature, and overheating reduction flow rate. and valve positions, and the like. Here, the coal supply amount of the boiler state measurement data may include the amount of fuel supplied to the boiler per hour. A British thermal unit (BTU) is a unit of energy representing boiler energy efficiency, and may include the amount of heat required to raise 1 pound of water from 60F to 61F by 1F. The primary air quantity may include an air quantity (ton/hour) supplied together with fuel for boiler combustion. The amount of secondary air may include an amount of excess air (ton/hour) that is additionally supplied to assist complete combustion of the fuel.

In addition, the fuel properties data include the type of coal, the proportion of coal of each type, calorific value (kcal/kg), total moisture (%), sulfur content (%), grindability (HGI: hardgrove grindability index), volatile matter (%), ash content (%) ), carbon (%), SiO2 (wt.%), CaO (wt.%), ignition temperature (°C), ash melting point (°C), and particle size (wt.%) may be included.

The boiler health monitoring apparatus 200 may configure time synchronization data obtained by synchronizing the boiler state measurement data, fuel property data, and power plant output data obtained from the coal-fired power plant 100 based on the fuel change time point.

The boiler health monitoring device 200 may generate training data to generate a power plant boiler health monitoring model. The boiler health monitoring apparatus 200 may generate and verify a power plant boiler health monitoring model by training a deep neural network model with the generated training data. The boiler health monitoring device 200 may use the power plant boiler health monitoring model to determine whether the boiler of the coal-fired power plant 100 is normal or abnormal, and if abnormal, may take a corresponding action.

The boiler health monitoring device 200 maintains a constant type of fuel in an output maintenance section in which power plant output data is constantly maintained from time synchronization data, and a combustion section of fuel included in the fuel property data, for generating training data. The output combined with the fuel holding section and the fuel holding section can be distinguished.

The boiler health monitoring apparatus 200 may determine the boiler health index according to the flow rate formation time based on the flow data of the reheater overheat reduction water included in the boiler state measurement data in the output and fuel maintenance section. Here, when the flow rate formation time based on the flow data of the reheater overheating reduction water does not exist when the boiler health index is determined, the boiler health index is determined as a normal state, and the flow rate formation time is the first critical time ( For example, if less than 5 minutes per hour), the boiler health indicator is determined as a first warning state, and when the flow rate forming time is greater than or equal to the first critical time and less than the second critical time (eg, 10 minutes per hour) The boiler health indicator may be determined as a second warning state, and when the flow rate formation time is greater than or equal to the second critical time, the boiler health indicator may be determined as a dangerous state.

Boiler health monitoring device 200, based on the boiler health index, with respect to a preset number (eg, 30) of boiler operation variables preselected to determine the operation state of the boiler among the power plant operation data, a preset cycle Training data including representative values of boiler operation variables calculated every (eg, 20 minutes) may be generated. Here, the preset number of boiler operation variables selected in advance may be data selected to determine the balance of heat absorption of the boiler. In addition, the training data may include a representative value of a boiler operation variable for each preset cycle as an input value, and may include a boiler health index in the corresponding cycle determined in the step of determining the boiler health index as a label.

In this embodiment, the boiler health monitoring device 200 may execute an artificial intelligence (AI) algorithm and/or a machine learning algorithm in a 5G communication environment to monitor the health of the boiler.

Here, artificial intelligence (AI) is a field of computer science and information technology that studies how computers can do the thinking, learning, and self-development that can be done with human intelligence. It could mean making it possible to imitate intelligent behavior.

In addition, artificial intelligence does not exist by itself, but is directly or indirectly related to other fields of computer science. In particular, in modern times, attempts are being made to introduce artificial intelligence elements in various fields of information technology and use them to solve problems in that field.

Machine learning is a branch of artificial intelligence, which can include fields of study that give computers the ability to learn without explicit programming. Specifically, machine learning can be said to be a technology to study and build a system and algorithms for learning based on empirical data, making predictions, and improving its own performance. Algorithms in machine learning can take the approach of building specific models to make predictions or decisions based on input data, rather than executing rigidly set static program instructions.

As a machine learning method of such an artificial neural network, both unsupervised learning and supervised learning can be used.

In addition, deep learning technology, which is a type of machine learning, can learn by going down to a deep level in multiple stages based on data. Deep learning can represent a set of machine learning algorithms that extract core data from a plurality of data as the level increases.

The deep learning structure may include an artificial neural network (ANN). For example, the deep learning structure is composed of a deep neural network (DNN) such as a convolutional neural network (CNN), a recurrent neural network (RNN), and a deep belief network (DBN). can be The deep learning structure according to the present embodiment may use various well-known structures. For example, the deep learning structure according to the present embodiment may include CNN, RNN, DBN, and the like. RNN is widely used in natural language processing, etc., and is an effective structure for processing time-series data that changes with time. DBN may include a deep learning structure composed of multi-layered restricted boltzman machine (RBM), a deep learning technique. By repeating RBM learning, when a certain number of layers is reached, a DBN having the corresponding number of layers can be configured. CNNs can include models that simulate human brain functions based on the assumption that when a person recognizes an object, it extracts the basic features of an object, then performs complex calculations in the brain and recognizes an object based on the result. .

On the other hand, learning of the artificial neural network can be accomplished by adjusting the weight of the connection line between nodes (and adjusting the bias value if necessary) so that a desired output is obtained for a given input. In addition, the artificial neural network may continuously update a weight value by learning. In addition, a method such as back propagation may be used for learning the artificial neural network.

In an embodiment of the present invention, the term 'machine learning' may be used interchangeably with the term 'machine learning'.

Regarding how to classify data in machine learning, many machine learning algorithms have been developed, such as decision trees, Bayesian networks, support vector machines (SVMs), and artificial neural networks ( artificial neural network (ANN) and the like.

The decision tree may include a decision method for performing classification and prediction by charting a decision rule in a tree structure. The Bayesian network may include a model expressing a probabilistic relationship (conditional independence) between a plurality of variables in a graph structure. Bayesian networks may be suitable for data mining through unsupervised learning. The support vector machine is a model of supervised learning for pattern recognition and data analysis, and can be mainly used for classification and regression analysis.

The artificial neural network models the operation principle of biological neurons and the connection relationship between neurons, and may represent an information processing system in which a number of neurons called nodes or processing elements are connected in the form of a layer structure. .

Artificial neural networks are models used in machine learning, and can include statistical learning algorithms inspired by neural networks in biology (especially the brain in the central nervous system of animals) in machine learning and cognitive science.

Specifically, the artificial neural network may refer to an overall model having problem-solving ability by changing the bonding strength of synapses through learning in which artificial neurons (nodes) that form a network by combining synapses.

The term artificial neural network may be used interchangeably with the term neural network.

The artificial neural network may include a plurality of layers, and each of the layers may include a plurality of neurons. Also, the artificial neural network may include neurons and synapses connecting neurons.

In general, artificial neural networks calculate the output value from the following three factors: (1) the connection pattern between neurons in different layers (2) the learning process to update the weight of the connection (3) the weighted sum of the input received from the previous layer It can be defined by the activation function it creates.

The artificial neural network may include network models such as a deep neural network (DNN), a recurrent neural network (RNN), a bidirectional recurrent deep neural network (BRDNN), a multilayer perceptron (MLP), and a convolutional neural network (CNN). , but is not limited thereto.

In this embodiment, the term 'layer' may be used interchangeably with the term 'layer'.

Artificial neural networks can be classified into single-layer neural networks and multi-layer neural networks according to the number of layers.

A general single-layer neural network may be composed of an input layer and an output layer. In addition, a general multilayer neural network may include an input layer, one or more hidden layers, and an output layer.

The input layer is a layer that receives external data. The number of neurons in the input layer is the same as the number of input variables, and the hidden layer is located between the input layer and the output layer. can The output layer may receive a signal from the hidden layer and output an output value based on the received signal. The input signal between neurons is multiplied by each connection strength (weight) and then summed. If the sum is greater than the threshold of the neuron, the neuron is activated and the output value obtained through the activation function can be output.

Meanwhile, a deep neural network including a plurality of hidden layers between an input layer and an output layer may be a representative artificial neural network that implements deep learning, which is a type of machine learning technology.

Meanwhile, the term 'deep learning' may be used interchangeably with the term 'deep learning'.

The artificial neural network may be trained using training data. Here, learning refers to a process of determining parameters of an artificial neural network using learning data to achieve objectives such as classification, regression, or clustering of input data. can As a representative example of parameters of an artificial neural network, a weight applied to a synapse or a bias applied to a neuron may be mentioned.

The artificial neural network learned by the training data may classify or cluster input data according to a pattern of the input data. On the other hand, an artificial neural network learned using training data may be called a trained model in this embodiment.

The following describes the learning method of the artificial neural network. A learning method of an artificial neural network may be largely classified into supervised learning, unsupervised learning, semi-supervised learning, and reinforcement learning.

Supervised learning can be a method of machine learning to infer a function from training data. And among these inferred functions, outputting continuous values is called regression, and predicting and outputting the class of the input vector is called classification.

In supervised learning, an artificial neural network can be trained in a state in which a label for training data is given. Here, the label may mean a correct answer (or result value) that the artificial neural network should infer when training data is input to the artificial neural network.

In this embodiment, when training data is input, the correct answer (or result value) to be inferred by the artificial neural network may be called a label or labeling data.

Also, in this embodiment, setting a label on the training data for learning of the artificial neural network may be referred to as labeling the labeling data on the training data. In this case, the training data and a label corresponding to the training data) constitute one training set, and may be input to the artificial neural network in the form of a training set.

On the other hand, training data represents a plurality of features, and labeling the training data may mean that the feature represented by the training data is labeled. In this case, the training data may represent the features of the input object in a vector form.

The artificial neural network may infer a function for the relationship between the training data and the labeling data by using the training data and the labeling data. In addition, parameters of the artificial neural network may be determined (optimized) through evaluation of the function inferred from the artificial neural network.

Unsupervised learning is a type of machine learning, in which the training data may not be labeled. Specifically, the unsupervised learning may be a learning method of learning the artificial neural network to find and classify patterns in the training data itself, rather than the association between the training data and the labels corresponding to the training data. Examples of unsupervised learning include clustering or independent component analysis.

In this embodiment, the term 'clustering' may be used interchangeably with the term 'clustering'.

Examples of artificial neural networks using unsupervised learning include a generative adversarial network (GAN) and an autoencoder (AE).

A generative adversarial neural network may be a machine learning method in which two different artificial intelligences, a generator and a discriminator, compete to improve performance. In this case, the generator is a model that creates new data, and can generate new data based on the original data. In addition, the discriminator is a model for recognizing patterns in data, and may play a role of discriminating whether input data is original data or new data generated by the generator. And the generator learns by receiving the data that did not deceive the discriminator, and the discriminator can learn by receiving the deceived data from the generator. Accordingly, the generator may evolve to deceive the discriminator as best as possible, and the discriminator may evolve to distinguish the original data and the data generated by the generator well.

An autoencoder can be a neural network that aims to reproduce the input itself as an output. The auto-encoder may include an input layer, at least one hidden layer, and an output layer. In this case, since the number of nodes in the hidden layer is smaller than the number of nodes in the input layer, the dimension of data is reduced, and thus compression or encoding may be performed. Also, data output from the hidden layer can enter the output layer. In this case, since the number of nodes in the output layer is greater than the number of nodes in the hidden layer, the dimension of data increases, and thus decompression or decoding may be performed.

On the other hand, the auto-encoder adjusts the neuron's connection strength through learning, so that input data can be expressed as hidden layer data. The hidden layer expresses information with fewer neurons than the input layer, and being able to reproduce the input data as an output may mean that the hidden layer found and expressed hidden patterns from the input data.

Semi-supervised learning is a type of machine learning, and may refer to a learning method using both labeled and unlabeled training data. As one of the techniques of semi-supervised learning, there is a technique of inferring a label of unlabeled training data and then performing learning using the inferred label. can

Reinforcement learning can include the theory that, given an environment where an agent can analyze what action should be taken at every moment, it can find the best path through experience without data. Reinforcement learning can be mainly performed by a markov decision process (MDP).

To explain the Markov decision process, firstly, an environment is given in which the information necessary for the agent to take the next action is given, secondly, how the agent will behave in that environment is defined, and thirdly, the agent is rewarded ( reward) and a penalty point for failure to do so, and fourthly, the optimal policy can be derived by repeating experiences until the future reward reaches the highest point.

The structure of an artificial neural network is specified by the model configuration, activation function, loss function or cost function, learning algorithm, optimization algorithm, etc., and hyperparameters are pre-trained. It is set, and then, a model parameter is set through learning and the content can be specified.

For example, factors determining the structure of an artificial neural network may include the number of hidden layers, the number of hidden nodes included in each hidden layer, an input feature vector, a target feature vector, and the like.

The hyperparameter may include several parameters to be initially set for learning, such as initial values of model parameters. And, the model parameter may include several parameters to be determined through learning.

For example, the hyperparameter may include an initial weight value between nodes, an initial bias value between nodes, a mini-batch size, a number of learning repetitions, a learning rate, and the like. In addition, the model parameters may include inter-node weights, inter-node biases, and the like.

The learning speed and accuracy of the artificial neural network may include features that largely depend on hyperparameters as well as the structure of the artificial neural network and the type of learning optimization algorithm. Therefore, in order to obtain a good learning model, it may be important not only to determine an appropriate artificial neural network structure and learning algorithm, but also to set appropriate hyperparameters.

In general, the hyperparameter can be set to an optimal value that provides a stable learning speed and accuracy as a result of training the artificial neural network by experimentally setting it to various values. If the above methods are used, the estimation of the state of the heating target can be more sophisticated.

The network 300 may serve to connect the coal-fired power plant 100 and the mixed coal combination determination device 200 . The network 300 includes wired networks such as local area networks (LANs), wide area networks (WANs), metropolitan area networks (MANs), and integrated service digital networks (ISDNs), wireless LANs, CDMA, Bluetooth, and satellite communication, for example. It may cover a wireless network such as, but the scope of the present invention is not limited thereto. Also, the network 300 may transmit/receive information using short-distance communication and/or long-distance communication. Here, the short-distance communication may include Bluetooth, radio frequency identification (RFID), infrared data association (IrDA), ultra-wideband (UWB), ZigBee, and wireless fidelity (Wi-Fi) technologies. Communication may include code division multiple access (CDMA), frequency division multiple access (FDMA), time division multiple access (TDMA), orthogonal frequency division multiple access (OFDMA), single carrier frequency division multiple access (SC-FDMA) technology. can

Network 300 may include connections of network elements such as hubs, bridges, routers, switches, and gateways. Network 300 may include one or more connected networks, eg, multiple network environments, including public networks such as the Internet and private networks such as secure enterprise private networks. Access to network 300 may be provided via one or more wired or wireless access networks. Furthermore, the network 300 may support an Internet of Things (IoT) network and/or 5G communication that exchanges and processes information between distributed components such as things.

FIG. 5 is a diagram schematically illustrating a configuration of a boiler health monitoring device including a training data generating device for training a power plant boiler health monitoring model in FIG. 3 . In the following description, descriptions of parts overlapping with those of FIGS. 1 to 4 will be omitted. Referring to FIG. 5 , the boiler health monitoring device 200 includes a communication unit 210 , a training data generation unit 220 , a model generation unit 230 , a model verification unit 240 , a memory 250 and a control unit 260 . may include In this embodiment, the training data generating unit 220 may be a training data generating device for training the health monitoring model of the power plant boiler in the claims to be described later. Also, in this embodiment, "unit" may be a hardware component such as a processor or circuit, and/or a software component executed by a hardware component such as a processor.

Referring to FIG. 5 , the communication unit 210 interworks with the network 300 to provide a communication interface necessary to provide a transmission/reception signal between the coal-fired power plant 100 and the boiler health monitoring device 200 in the form of packet data. there is. In addition, the communication unit 210 transmits a predetermined information request signal from the boiler health monitoring device 200 to the coal-fired power plant 100, and receives the response data processed by the coal-fired power plant 100, the boiler health monitoring device ( 200) can be transmitted. In addition, the communication unit 210 may be a device including hardware and software necessary for transmitting and receiving signals such as control signals or data signals through wired/wireless connection with other network devices.

In addition, the communication unit 210 may support various things intelligent communication (Internet of things (IoT), Internet of everything (IoE), Internet of small things (IoST), etc.), and M2M (machine to machine) communication, V2X ( Vehicle to everything communication) communication, D2D (device to device) communication, etc. may be supported.

The training data generation unit 220 determines the boiler health index using the boiler state measurement data obtained from the coal-fired power plant 100 through the communication unit 210, the fuel property data, and the power plant output data, and the boiler health index training data can be generated based on A detailed description of the training data generator 220 will be described later.

The model generator 230 may generate a power plant boiler health monitoring model by training a deep neural network model with the training data generated by the training data generator 220 . Here, the deep neural network model can be trained by a supervised learning method or an unsupervised learning method. In the case of the supervised learning method, training data in a steady state and training data in an abnormal state are provided as a deep neural network model, so that the deep neural network model can train the criteria for determining whether a boiler is normal in a power plant. In the case of the unsupervised learning method, only training data in a steady state is provided as a deep neural network model, and after learning a pattern in a steady state, when non-normal data is input, it can be determined as an abnormal state.

The model verification unit 240 may input verification data not used for training into the power plant boiler health monitoring model to check the performance of detecting abnormalities in the power plant boiler.

The memory 250 may store various data necessary for the operation of the boiler health monitoring apparatus 200 . In this embodiment, the memory 250 may be provided inside the control unit 260 or may be provided outside the control unit 260 . In addition, the memory 250 may store the data mart configuration information generated by the training data generator 220 , the analysis mart configuration information, the training data, and the power plant boiler health monitoring model generated by the model generator 230 . .

In addition, the memory 250 is a command to be executed by the training data generator 220 under the control of the boiler health monitoring device 200, for example, boiler state measurement data from the coal-fired power plant 100, fuel properties data, and power plant A command for acquiring power plant operation data including output data, an output maintenance section in which power plant output data is kept constant, and a fuel maintenance section in which the type of fuel is kept constant among the combustion sections of fuel included in the fuel property data are combined. A command to distinguish the output and fuel maintenance section, a command to determine the boiler health index according to the flow rate formation time based on the flow data of the reheater overheat reduction water included in the boiler status measurement data, in the output and fuel maintenance section, power plant For a preset number of boiler operation variables pre-selected to determine the operation state of the boiler among the operation data, based on the boiler health index, training data including representative values of the boiler operation variables calculated for each preset cycle is generated. You can store commands, etc.

Here, the memory 250 may include magnetic storage media or flash storage media, but the scope of the present invention is not limited thereto. Such memory 250 may include internal memory and/or external memory, volatile memory such as DRAM, SRAM, or SDRAM, one time programmable ROM (OTPROM), PROM, EPROM, EEPROM, mask ROM, flash ROM, Non-volatile memory, such as NAND flash memory, or NOR flash memory, SSD. It may include a flash drive such as a compact flash (CF) card, an SD card, a Micro-SD card, a Mini-SD card, an Xd card, or a memory stick, or a storage device such as an HDD.

The controller 260 may control the overall operation of the boiler health monitoring device 200 . Here, the controller 260 may include all kinds of devices capable of processing data, such as a processor. Here, the 'processor' may refer to, for example, a data processing device embedded in hardware having a physically structured circuit to perform a function expressed as a code or command included in a program. As an example of the data processing apparatus embedded in the hardware as described above, a microprocessor, a central processing unit (CPU), a processor core, a multiprocessor, an application-specific integrated (ASIC) circuit) and a processing device such as a field programmable gate array (FPGA), but the scope of the present invention is not limited thereto.

In this embodiment, the control unit 260 is an artificial intelligence (AI) algorithm and/or machine learning in a 5G communication environment so that the boiler health monitoring device 200 can monitor the health of the power plant boiler in an optimal state. (machine learning) algorithms can be executed.

6 is a diagram schematically illustrating data obtained by the training data generating device of FIG. 5 from a power plant, and FIG. 7 is an output and fuel maintenance section extracted from the data collected by the training data generating device of FIG. 5 FIG. 8 is a diagram for schematically explaining the criteria for determining the health of the boiler determined by the training data generating device in FIG. 5, and FIG. 9 is the training data generating device in FIG. 5 It is a diagram illustrating schematically the boiler operation variables selected and processed for generating training data, and FIG. 10 is a diagram illustrating schematically the final analysis mart determined by the training data generating device in FIG. 5 . In the following description, descriptions of parts overlapping with those of FIGS. 1 to 5 will be omitted.

In this embodiment, the training data generation unit 220 includes an acquisition unit 221 , a preprocessing unit 222 , a data mart configuration unit 223 , a determination unit 224 , a generation unit 225 , and an analysis mart configuration unit 226 . ) may be included.

The acquisition unit 221 may acquire the power plant operation data including the boiler state measurement data measured by the sensor, the fuel property data, and the power plant output data from the coal-fired power plant 100 through the communication unit 210 .

6 illustrates an example of data obtained by the acquisition unit 221 from the coal-fired power plant 100 through the communication unit 210 . Referring to Fig. 6, Fig. 6a shows the power plant output data, Fig. 6b shows the amount of coal feed, Fig. 6c shows the final reheater A/B (first valve/second valve) outlet temperature, and Fig. 6d shows the ash It can represent the flow rate of hot air overheating reduction water. In addition, a vertical dotted line from FIGS. 6A to 6D may indicate a fuel change time (ammunition type change time).

The preprocessor 222 may process the boiler state measurement data, the fuel property data, and the power plant output data obtained by the acquisition unit 221 in a form suitable for training. Since the boiler state measurement data, fuel property data, and power plant output data acquired by the acquisition unit 221 have different units and ranges for each factor, it is necessary to normalize them to a certain range. Accordingly, the preprocessor 222 may normalize the boiler state measurement data, the fuel property data, and the power plant output data to a certain range.

If, when the training data of the power plant boiler health monitoring model is out of the minimum and maximum values of the existing learning data, when the minimum-maximum normalization is applied, the preprocessor 222 may deviate from the normalization range. The data can be automatically adjusted to the minimum/maximum values of the existing training data.

The preprocessor 222 may include a missing value deletion filter and/or a steady state detection filter. If missing values exist in the input data, the missing value deletion filter can delete the information so that it cannot be used in the power plant boiler health monitoring model. The steady state data filter is an exponentially weighted moving average filter to receive only the stable state data among the boiler state measurement data obtained from the coal-fired power plant 100, fuel properties data, and power plant output data. By adopting it, it is possible to determine the stable state of the data obtained from the coal-fired power plant 100 . The exponential weighted moving average filter may be determined to be in a stable state when the filter error is less than or equal to a predetermined filter error value during a specific period T.

The data mart configuration unit 223 may be processed into time synchronization data as a result of integrating and synchronizing boiler state measurement data, fuel property data, and power plant operation data in a stable state processed by the pre-processing unit 222 . The processed time synchronization data may be stored in an analysis mart (not shown) provided in the data mart configuration unit 223 .

The data mart configuration unit 223 may integrate the boiler state measurement data, fuel property data, and power plant output data, and configure time synchronization data synchronized based on the fuel change time.

The data mart configuration unit 223 is configured to correct the boiler state measurement data and power plant output data as time series data acquired at a preset time interval (eg, 10 minutes), and 8 hours delay from the time of the fuel phase coal combustion section of the same type of coal. It is possible to integrate fuel properties data as static data with Thereafter, the data mart configuration unit 223 may configure time synchronization data in which the boiler state measurement data, fuel property data, and power plant output data are synchronized based on the fuel change time, and may be stored in the analysis mart. Here, the term "sangtan" may be used to include not only the act of inserting fuel but also the act of changing fuel.

The fuel properties data may include information on properties of fuels that are periodically replaced. Since the coal silo can be burned for about 8 hours once it is full, the coal-fired power plant 100 can load fuel into the coal silo three times a day. At this time, depending on the fuel stock and combustion conditions, the same type of ammunition may be continuously loaded, or the type of ammunition may be changed at any time to burn. Meanwhile, in the present embodiment, the low carbon tank may refer to a space containing only coal, as well as a space containing other types of fuel.

In addition, the fuel change time may indicate the time of combustion by changing to a different type of coal in general, and in the data format, which type of ammunition is loaded into the low coal tank in units of 8 hours may be recorded. That is, the coal-fired power plant 100 determines that the fuel change has not been made when the fuel loaded in the low carbon tank has the same properties as the previously loaded fuel, and the fuel loaded in the low carbon tank has a different property from the previously loaded fuel. It can be determined that a fuel change has been made.

When the data mart configuration unit 223 stores the time synchronization data in the analysis mart, the analysis mart after processing the combustion start time and the combustion end time of the fuel from the time when the fuel is loaded in the low coal tank included in the fuel property data. can be stored in

The data mart configuration unit 223 may extract a combustion start time of the fuel based on the coal phase time. The data mart configuration unit 223 adds a preset time required for complete combustion of the fuel to the combustion start time of the fuel when the boiler operates at the maximum output in a state in which the fuel tank is full of fuel, and the combustion end time of the fuel can be processed into Here, from the combustion start time of the fuel to the combustion end time of the fuel can be specified as the combustion section of the fuel.

The time recorded in the fuel property information is generally the time the fuel is loaded into the low carbon tank, but it may take time to actually burn inside the boiler. Therefore, the time the fuel is loaded into the low carbon tank can be used as the combustion start time. In addition, because it is difficult to know when the fuel loaded in the low coal tank is burned due to the circumstances of the coal-fired power plant 100, in general, when the boiler is operated at maximum output in a state in which the low coal tank is full, it takes 8 hours to burn all of it. , the time the fuel is loaded into the low carbon tank, that is, the time corrected by adding 8 hours to the combustion start time may be used as the combustion end time. For example, if fuel is loaded into the low carbon tank at 00:00 on March 15, 2020, it is processed as the combustion start time, and March 15, 2020 08:00, which is 8 hours added to the combustion start time, is set as the combustion end time. can be processed

In addition, the data mart configuration unit 223 temporarily expands the boiler state variability at the time of the load-transient-phase or fuel change of the power plant output data in order to extract characteristic information necessary for learning the power plant boiler health monitoring model. Additional data purification may be performed to reduce the risk of false alarms that may occur due to this.

The data mart configuration unit 223 outputs a combination of an output maintenance section in which the power plant output data is constantly maintained from the time synchronization data and a fuel maintenance section in which the type of fuel is kept constant in a combustion section of fuel included in the fuel property data. And the fuel maintenance section may be stored separately. Here, the fuel combustion section may be data indicating the type of fuel being burned for each time. Also, the output and fuel maintenance section may be a section in which the type of fuel is maintained while the output is maintained. Furthermore, dividing the output and fuel maintenance period may include finding a time period in which both output and fuel types are maintained.

The data mart configuration unit 223 applies a change point detection algorithm to the power plant output data, and includes an output maintenance section in which the power plant output data is kept constant, and an output change in which the power plant output data changes upward or downward. section can be extracted.

7 shows an example in which the data mart configuration unit 223 divides the output and fuel maintenance sections. Referring to FIG. 7 , by applying a change point detection algorithm to the power plant output data, the output maintenance section and the output change section are extracted, and the output and the fuel maintenance section are divided by integrating with the fuel (fuel) combustion history included in the fuel property data. can

The determination unit 224 may determine a boiler health index with respect to the boiler state measurement data. Here, the determination of the boiler health index with respect to the boiler state measurement data means, for example, the boiler state measurement data values (representative values of each variable during the corresponding period - for example, average values) obtained during the period in which the health is in the normal state ) may include the meaning of labeling as normal for a set of .

In this embodiment, the soundness determination index (standard) of the boiler in the power plant may be whether the reheater overheat reduction unit operates. Since the overheating reduced water flow rate of the reheater is formed due to the heat absorption imbalance of the boiler feedwater/steam system, resulting in loss of boiler efficiency, the condition causing the loss of the boiler can be defined as an abnormal state.

The determination unit 224 is based on the position data of the reheater overheating reduction water A/B valve (the first valve and the second valve) included in the boiler state measurement data, at least one of the first valve and the second valve When the flow rate is formed by opening more than a preset ratio (for example, 1%), it can be determined that an abnormal state has occurred due to an imbalance in the heat absorption of the boiler, and based on this, it is possible to determine the boiler health index.

The determination unit 224 determines the hourly flow rate based on the flow rate data of the reheater overheat reduction water included in the boiler state measurement data for a section in which the operation time is maintained for 30 minutes or more among the output and fuel maintenance sections, And/or based on the position data of the reheater overheating reduction first valve and the second valve included in the boiler state measurement data, at least one of the first valve and the second valve is opened by a predetermined ratio or more. Depending on the hourly flow rate formation time, the boiler health index can be determined.

FIG. 8 shows the criteria for determining the health of the boiler determined by the determining unit 224 . Referring to FIG. 8 , when the flow rate formation time does not exist, the determination unit 224 may determine the boiler health index as a steady state. When the flow rate formation time is less than the first critical time (eg, 5 minutes per hour), the determination unit 224 may determine the boiler health index as the first attention state. The determination unit 224 may determine the boiler health index as the second caution state when the flow rate formation time is equal to or longer than the first critical time and less than the second critical time (eg, 10 minutes per hour). The determination unit 224 may determine the boiler health index as a dangerous state when the flow rate formation time is equal to or greater than the second threshold time.

Since overheating reduction operation leads to a decrease in boiler efficiency, it is necessary to select an appropriate boiler operation variable to comprehensively determine the operation status of the boiler, and to extract features through the creation of processing variables.

The generator 225 may select and process a preset number of boiler operation variables in order to determine the operation state of the boiler from among the power plant operation data. Here, the boiler operation variable may be data selected to determine the balance of heat absorption of the boiler. The generator 225 may generate a boiler operation variable including the power plant output, the inlet temperature and outlet temperature of the heat transfer unit of the water supply and steam system, and the temperature change of the heat transfer unit inlet temperature and the outlet temperature of the water supply and steam system. The generator 225 may select and process boiler operation variables to comprehensively express the thermal balance of the boiler feedwater/steam system, which is a direct cause of boiler abnormal conditions. The generator 225 may select an inlet/outlet temperature tag distinguishable in the boiler feedwater/steam system process from the boiler measurement data as a boiler operation variable, and may select an inlet/outlet temperature change amount of a main part as a boiler operation processing variable.

9 shows the boiler operation variables generated by selection and processing by the generator 225 . 9, the selected boiler operation variables are 1-MW power plant output (M/W), 2- PE_IH_AMT Primary Economizer inlet header temperature (℃), 3-SE_O_AST Second Economizer outlet header temperature (℃), 4-ST_O1_ST furnace Water-cooled wall water temperature (℃), 5- ST_O3_ST Furnace water-cooled wall water supply temperature (℃), 6-ST_O5_ST Furnace water-cooled wall water supply temperature (℃), 7-SP_O_AST Water separator outlet header water supply temperature (℃), 8-DSH_IL1_ST Division Superheater Inlet Header Steam Temperature (°C), 9-DSH_O1_ST Division Superheater Outlet Header Steam Temperature (°C), 10-PSH_IA_AST Platen Superheater A Inlet Header Steam Temperature (°C), 11-PSH_OA_AT Platen Superheater A Outlet Header Steam Temperature (°C), 12 -PSH_IB_AST Platen Superheater B Inlet Header Vapor Temperature (°C), 13-PSH_OB_AT Platen Superheater B Outlet Header Vapor Temperature (°C), 14-FSH_IA_ST Final Superheater A Inlet Header Vapor Temperature (°C), 15-FSH_OHA_AST Final Superheater A Outlet Header Vapor Temp (℃), 16-FSH_IB_AST Final Superheater B Inlet Header Vapor Temp (℃), 17-FSH_OHB_AST Final Superheater B Outlet Header Vapor Temp (℃), 18-HR_IA_AT Final reheater A Inlet Header Vapor Temp (℃), 19-HR_OA_AT final reheater A outlet header steam temperature (℃), 20-HR_IB_AT final reheater B inlet header steam temperature (℃), 21-HR_OB_AT final reheater B outlet header steam temperature (℃), 22-D_EC Processing parameters: (3) - ( 2) Economizer temperature change, 23-D_WW processing variable (7) - (3) Water cooling wall temperature change, 24-D_DSH processing variable: (9) - (8) Division superheater temperature change, 25-D_PSHA process variable: (11) - (10) Platen A superheater temperature change, 26-D_PSHB process variable: (13) - (12) Platen B superheater temperature change, 27-D_FSHA Process variable: (15) - (14) Final Superheater A temperature change, 28-D_FSHB Process variable: (17) - (16) Fianl Superheater B temperature change, 29-D_HRA Process variable: (19) - (18) Final material Hot air A temperature change, and 30-D_HRB processing parameters: (21) - (20) final reheater B temperature change.

The generator 225 may standardize the values of all boiler operation variables into a preset section (0, 1) by executing min-max normalization on the selected and processed boiler operating variables. there is. This may be to prevent the deterioration of the learning performance of the artificial intelligence model due to the difference in scale between boiler operation variables.

The generator 225 includes a representative value of the boiler operation variables calculated for each predetermined cycle (eg, 20 minutes) based on the boiler health index for a predetermined number of selected and processed boiler operation variables. You can create training data. Here, the representative value may be selected to represent the value of the driving variable of the corresponding section, such as an average value, a median value, and a mode value. In addition, the training data includes the representative value of the boiler operation variable for each preset cycle as an input value, and the boiler health index (normal state, first attention state, second One of caution state and danger state) can be included as a label.

The analysis mart configuration unit 226 is configured such that the position of the reheater overheating reduction water A/B valve (the first valve and the second valve) is a preset ratio (for example, 1 %) Boiler health indicators (normal state, first caution state, second caution state, and dangerous state) determined based on the time when the flow rate was formed due to abnormal opening, boiler operation status, and heat balance of the water supply/medium system. By selecting and processing the number of boiler operation variables (for example, 30), a final analysis mart may be configured and stored with representative values calculated for each preset cycle (for example, 20 minutes). 10 shows the final analysis mart configured by the analysis mart configuration unit 226.

11 is a diagram schematically illustrating the health detection and boiler health monitoring model learning and verification performed by the boiler health monitoring device in FIG. 3 A description thereof will be omitted.

Referring to FIG. 11 , in the present embodiment, the presence or absence of operation (flow rate formation time) of the overheat reduction water caused by overheating of the reheater heat transfer unit due to the imbalance of heat absorption in the boiler may be an indicator of boiler health.

The boiler health monitoring apparatus 200 may generate a power plant boiler health monitoring model by training a deep neural network model with the training data generated by the training data generator 220 . Here, the deep neural network model can be trained by a supervised learning method or an unsupervised learning method. In the case of the supervised learning method, training data in a steady state and training data in an abnormal state are provided as a deep neural network model, so that the deep neural network model can train the criteria for determining whether a boiler is normal in a power plant. In the case of the unsupervised learning method, only training data in a steady state is provided as a deep neural network model, and after learning a pattern in a steady state, when non-normal data is input, it can be determined as an abnormal state.

In this embodiment, supervised learning includes a support vector machine, a K-nearest neighbor (K-NN), an ensemble K-NN, and a decision tree decision tree) can be used. In this case, training performance can be maximized by dividing the training data into 5 parts and performing cross-validation during the model learning process. For unsupervised learning, an autoencoder widely used for anomaly detection can be used as a representative deep network-based unsupervised learning model.

The data for each state in the analysis mart is divided into training data (75%) and validation data (25%) at a 3:1 ratio, and an artificial intelligence model to maximize anomaly detection performance based on the training data The process of optimizing the hyperparameters of At this time, the supervised learning model detects abnormalities in the power plant boiler in a way that classifies normal/abnormal (first caution, second caution, danger), and the unsupervised learning model determines whether the input data follows the pattern of the normal state. By determining whether or not it is not, it is possible to detect abnormalities in the power plant boiler. In order to verify the actual performance of the trained power plant boiler health monitoring model, verification data not used for training is input to the power plant boiler health monitoring model to check the performance of detecting abnormalities in the power plant boiler.

12 is a flowchart for explaining a training data generation method for training a health monitoring model of a power plant boiler according to the present embodiment. In the following description, descriptions of parts overlapping with those of FIGS. 1 to 11 will be omitted.

Referring to FIG. 12 , in step S1210 , the training data generating unit 220 acquires power plant operation data including boiler state measurement data, fuel property data, and power plant output data from the coal-fired power plant 100 .

Thereafter, the training data generator 220 may configure time synchronization data in which the boiler state measurement data, fuel property data, and power plant output data are synchronized based on the fuel change time. When configuring time synchronization data, the training data generator 220 may process a combustion start time and a combustion end time of the fuel from the time of loading the fuel in the low coal tank included in the fuel property data. The training data generator 220 extracts the combustion start time of the fuel based on the coal time at which the fuel is loaded into the low coal tank included in the fuel property data, and when the boiler operates at maximum output in a state in which the low coal tank is full of fuel , can be processed as the combustion end time of the fuel by adding the preset time required for the complete combustion of the fuel to the combustion start time of the fuel. can be specified.

In step S1220, the training data generator 220 combines the output maintenance section in which the power plant output data is constantly maintained and the fuel maintenance section in which the type of fuel is kept constant among the combustion sections of the fuel included in the fuel property data. and a fuel maintenance section. The training data generation unit 220 applies a change point detection algorithm to the power plant output data, an output maintenance section in which the power plant output data is kept constant, and an output change in which the power plant output data changes upward or downward. section can be extracted.

In step S1230, the training data generator 220 determines the boiler health index according to the flow rate formation time based on the flow data of the reheater overheat reduction water included in the boiler state measurement data in the output and fuel maintenance section. Here, the flow rate formation time for the reheater overheat reduction water is based on the position data of the reheater overheat reduction water first valve and the second valve included in the boiler state measurement data, at least one of the first valve and the second valve It may be formed as the opening degree exceeds this preset ratio.

The training data generator 220 may determine the boiler health index as a steady state as the flow rate formation time does not exist. The training data generator 220 may determine the boiler health index as the first attention state according to the flow rate formation time being less than the first threshold time. The training data generation unit 220 may determine the boiler health index as the second attention state according to the flow rate formation time being greater than or equal to the first threshold time and less than the second threshold time. The training data generation unit 220 may determine the boiler health index as a dangerous state as the flow rate formation time is equal to or greater than the second threshold time.

In step S1240, the training data generator 220 calculates the boiler for each preset cycle based on the boiler health index for a preset number of boiler operation variables selected in advance to determine the operating state of the boiler among the power plant operation data. Generate training data including representative values of driving variables.

The training data generator 220 may select and process a preset number of boiler operation variables in order to determine the operation state of the boiler from among the power plant operation data. Here, the boiler operation variable may be data selected to determine the balance of heat absorption of the boiler. The training data generation unit 220 may generate a boiler operation variable including the temperature change of the power plant output, the inlet temperature and outlet temperature of the heat transfer unit of the water supply and steam system, and the inlet temperature and the outlet temperature of the heat transfer unit of the water supply and steam system. . Here, the representative value may be selected to represent the value of the driving variable of the corresponding section, such as an average value, a median value, and a mode value. In addition, the training data includes the representative value of the boiler operation variable for each preset cycle as an input value, and the boiler health index (normal state, first attention state, second One of caution state and danger state) can be included as a label.

Thereafter, the boiler health monitoring device 200 including the training data generator 220 may generate and verify the power plant boiler health monitoring model by training the deep neural network model with the training data. The boiler health monitoring device 200 may use the power plant boiler health monitoring model to determine whether the boiler of the coal-fired power plant 100 is normal or abnormal, and if abnormal, may take a corresponding action.

The above-described embodiment according to the present invention may be implemented in the form of a computer program that can be executed through various components on a computer, and such a computer program may be recorded in a computer-readable medium. In this case, the medium includes a hard disk, a magnetic medium such as a floppy disk and a magnetic tape, an optical recording medium such as CD-ROM and DVD, a magneto-optical medium such as a floppy disk, and a ROM. , RAM, flash memory, and the like, and hardware devices specially configured to store and execute program instructions.

Meanwhile, the computer program may be specially designed and configured for the present invention, or may be known and used by those skilled in the computer software field. Examples of the computer program may include not only machine code generated by a compiler, but also high-level language code that can be executed by a computer using an interpreter or the like.

In the specification of the present invention (especially in the claims), the use of the term "above" and similar referential terms may be used in both the singular and the plural. In addition, when a range is described in the present invention, each individual value constituting the range is described in the detailed description of the invention as including the invention to which individual values belonging to the range are applied (unless there is a description to the contrary). same as

The steps constituting the method according to the present invention may be performed in an appropriate order unless the order is explicitly stated or there is no description to the contrary. The present invention is not necessarily limited to the order in which the steps are described. The use of all examples or exemplary terms (eg, etc.) in the present invention is merely for the purpose of describing the present invention in detail, and the scope of the present invention is limited by the examples or exemplary terms unless limited by the claims. it is not going to be In addition, those skilled in the art will appreciate that various modifications, combinations, and changes may be made in accordance with design conditions and factors within the scope of the appended claims or their equivalents.

Therefore, the spirit of the present invention should not be limited to the above-described embodiments, and the scope of the spirit of the present invention is not limited to the scope of the scope of the present invention. will be said to belong to

100: coal-fired power plant
200: boiler health monitoring device
300: network

Claims (20)

A training data generating method for training the health monitoring model of a power plant boiler, wherein each step is performed by a training data generating device for training the health monitoring model of the power plant boiler,
acquiring power plant operation data including boiler state measurement data measured by a sensor from a power plant, fuel properties data, and power plant output data;
Separating an output and fuel maintenance section combining an output maintenance section in which the power plant output data is constantly maintained and a fuel maintenance section in which the type of fuel is maintained constant among the combustion sections of the fuel included in the fuel property data ;
determining a boiler health index according to a flow rate formation time based on flow rate data of a reheater overheat reduction water included in the boiler state measurement data in the output and fuel maintenance section; and
With respect to a preset number of boiler operation variables pre-selected to determine the operating state of the boiler among the power plant operation data, the boiler operation variable includes a representative value calculated for each predetermined cycle based on the boiler health index generating training data;
The training data includes a representative value of a boiler operation variable for each preset cycle as an input value, and includes a boiler health index in the corresponding cycle determined in the step of determining the boiler health index as a label,
A method of generating training data to train a health monitoring model of a power plant boiler. The method of claim 1,
Before the step of dividing the output and fuel maintenance section,
The method further comprising the step of configuring time synchronization data in which the boiler state measurement data, the fuel property data, and the power plant output data are synchronized based on a fuel change time point,
A method of generating training data to train a health monitoring model of a power plant boiler. 3. The method of claim 2,
The step of configuring the time synchronization data includes:
Further comprising the step of processing the combustion start time and the combustion end time of the fuel from the phase coal time when fuel is loaded in the low carbon tank included in the fuel property data,
A method of generating training data to train a health monitoring model of a power plant boiler. 4. The method of claim 3,
The processing step is
extracting the combustion start time of the fuel based on the coal phase time at which the fuel is loaded in the low coal tank included in the fuel property data; and
When the boiler operates at maximum output in a state in which the fuel tank is full, the preset time required for the fuel to completely burn is added to the combustion start time of the fuel and processed as the combustion end time of the fuel comprising steps;
To specify from the combustion start time of the fuel to the combustion end time of the fuel as the combustion section of the fuel,
A method of generating training data to train a health monitoring model of a power plant boiler. The method of claim 1,
The step of dividing the output and fuel maintenance section is,
applying a change point detection algorithm to the power plant output data to extract an output maintenance section in which the power plant output data is kept constant and an output change section in which the power plant output data changes upward or downward containing,
A method of generating training data to train a health monitoring model of a power plant boiler. The method of claim 1,
The step of determining the boiler health index is,
determining the boiler health index as a steady state according to the absence of the flow rate formation time;
determining the boiler health index as a first alert state according to the flow rate formation time being less than a first threshold time;
determining the boiler health index as a second alert state according to the flow rate formation time being greater than or equal to a first threshold time and less than a second threshold time; and
Comprising the step of determining the boiler health indicator as a dangerous state according to the flow rate formation time is equal to or greater than a second threshold time,
A method of generating training data to train a health monitoring model of a power plant boiler. The method of claim 1,
In the step of determining the boiler health index,
The flow rate formation time for the reheater overheat reduction water is based on the position data of the reheater overheat reduction water first and second valves included in the boiler state measurement data, the first valve and the second valve Estimated as one or more of them are opened more than a predetermined ratio,
A method of generating training data to train a health monitoring model of a power plant boiler. The method of claim 1,
In the step of generating the training data,
The pre-selected preset number of boiler operation variables is data selected to determine the balance of heat absorption of the boiler,
A method of generating training data to train a health monitoring model of a power plant boiler. The method of claim 1,
In the step of generating the training data,
The pre-selected preset number of boiler operation variables include power plant output, the inlet temperature and outlet temperature of the heat transfer part of the feedwater and steam system, and the temperature change of the heat transfer part inlet temperature and the outlet temperature of the water supply and steam system,
A method of generating training data to train a health monitoring model of a power plant boiler. The method of claim 1,
In the step of generating the training data,
The method further comprising the step of normalizing the pre-selected preset number of boiler operation variables to a preset section by executing minimum-maximum normalization,
A method of generating training data to train a health monitoring model of a power plant boiler. A computer-readable recording medium storing a computer program for executing the method of any one of claims 1 to 10 using a computer. A training data generating device for training a health monitoring model of a power plant boiler, comprising:
an acquisition unit for acquiring power plant operation data including boiler state measurement data measured by a sensor from a power plant, fuel property data, and power plant output data;
Data for separating an output and fuel maintenance section combining an output maintenance section in which the power plant output data is constantly maintained and a fuel maintenance section in which the type of fuel is maintained constant among the combustion sections of the fuel included in the fuel property data mart composition unit;
a determination unit configured to determine a boiler health index according to a flow rate formation time based on flow data of a reheater overheat reduction water included in the boiler state measurement data in the output and fuel maintenance section; and
With respect to a preset number of boiler operation variables pre-selected to determine the operating state of the boiler among the power plant operation data, the boiler operation variable includes a representative value calculated for each predetermined cycle based on the boiler health index a generator for generating training data;
The training data includes a representative value of a boiler operation variable for each preset cycle as an input value, and includes a boiler health index in the corresponding cycle determined by the determination unit as a label,
A training data generating device for training the health monitoring model of a power plant boiler. 13. The method of claim 12,
The data mart component,
Prior to dividing the output and fuel maintenance section, the boiler state measurement data, the fuel property data, and the power plant output data to configure time synchronization data synchronized with the fuel change time point,
A training data generating device for training the health monitoring model of a power plant boiler. 14. The method of claim 13,
The data mart component,
When configuring the time synchronization data, the combustion start time of the fuel is extracted based on the coal time at which the fuel is loaded in the low coal tank included in the fuel property data,
When the boiler operates at maximum output in a state in which the fuel tank is full, the preset time required for the fuel to be completely burned is added to the combustion start time of the fuel and processed as the combustion end time of the fuel becomes,
To specify from the combustion start time of the fuel to the combustion end time of the fuel as the combustion section of the fuel,
A training data generating device for training the health monitoring model of a power plant boiler. 13. The method of claim 12,
The data mart component,
A change point detection algorithm is applied to the power plant output data to extract an output maintenance section in which the power plant output data is kept constant and an output change section in which the power plant output data changes upward or downward. felled,
A training data generating device for training the health monitoring model of a power plant boiler. 13. The method of claim 12,
The determining unit is
determining the boiler health index as a steady state according to the absence of the flow rate formation time;
determining the boiler health index as a first alert state according to the flow rate formation time being less than a first threshold time;
determining the boiler health index as a second alert state according to the flow rate formation time being greater than or equal to a first threshold time and less than a second threshold time; and
Comprising the step of determining the boiler health indicator as a dangerous state according to the flow rate formation time is equal to or greater than a second threshold time,
A training data generating device for training the health monitoring model of a power plant boiler. 13. The method of claim 12,
In the decision unit,
The flow rate formation time for the reheater overheat reduction water is based on the position data of the reheater overheat reduction water first and second valves included in the boiler state measurement data, the first valve and the second valve Estimated as one or more of them are opened more than a predetermined ratio,
A training data generating device for training the health monitoring model of a power plant boiler. 13. The method of claim 12,
In the generator,
The pre-selected preset number of boiler operation variables is data selected to determine the balance of heat absorption of the boiler,
A training data generating device for training the health monitoring model of a power plant boiler. 13. The method of claim 12,
In the generator,
The pre-selected preset number of boiler operation variables include power plant output, the inlet temperature and outlet temperature of the heat transfer part of the feedwater and steam system, and the temperature change of the heat transfer part inlet temperature and the outlet temperature of the water supply and steam system,
A training data generating device for training the health monitoring model of a power plant boiler. 13. The method of claim 12,
The generating unit,
configured to normalize to a preset section by executing min-maximum normalization for the pre-selected preset number of boiler operation variables,
A training data generating device for training the health monitoring model of a power plant boiler.

Images