CN117471906A - Automatic control method for reducing emission of mixed-combustion multi-source biomass smoke of coal-fired boiler - Google Patents
Automatic control method for reducing emission of mixed-combustion multi-source biomass smoke of coal-fired boiler Download PDFInfo
- Publication number
- CN117471906A CN117471906A CN202311263477.4A CN202311263477A CN117471906A CN 117471906 A CN117471906 A CN 117471906A CN 202311263477 A CN202311263477 A CN 202311263477A CN 117471906 A CN117471906 A CN 117471906A
- Authority
- CN
- China
- Prior art keywords
- data
- dnn
- coal
- flue gas
- lstm
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 35
- 239000002028 Biomass Substances 0.000 title claims abstract description 29
- 238000002485 combustion reaction Methods 0.000 title claims abstract description 27
- 239000000779 smoke Substances 0.000 title claims abstract description 16
- 238000012549 training Methods 0.000 claims abstract description 29
- 230000004927 fusion Effects 0.000 claims abstract description 23
- UGFAIRIUMAVXCW-UHFFFAOYSA-N Carbon monoxide Chemical compound [O+]#[C-] UGFAIRIUMAVXCW-UHFFFAOYSA-N 0.000 claims abstract description 22
- 239000003546 flue gas Substances 0.000 claims abstract description 22
- 239000003795 chemical substances by application Substances 0.000 claims abstract description 18
- 239000003344 environmental pollutant Substances 0.000 claims abstract description 17
- 231100000719 pollutant Toxicity 0.000 claims abstract description 17
- 238000011217 control strategy Methods 0.000 claims abstract description 15
- 238000002156 mixing Methods 0.000 claims abstract description 15
- 239000003245 coal Substances 0.000 claims abstract description 12
- 238000012545 processing Methods 0.000 claims abstract description 10
- 238000013528 artificial neural network Methods 0.000 claims description 25
- 238000004422 calculation algorithm Methods 0.000 claims description 20
- 239000000446 fuel Substances 0.000 claims description 13
- 210000002569 neuron Anatomy 0.000 claims description 10
- QVGXLLKOCUKJST-UHFFFAOYSA-N atomic oxygen Chemical compound [O] QVGXLLKOCUKJST-UHFFFAOYSA-N 0.000 claims description 9
- 230000001276 controlling effect Effects 0.000 claims description 9
- 239000001301 oxygen Substances 0.000 claims description 9
- 229910052760 oxygen Inorganic materials 0.000 claims description 9
- 238000005457 optimization Methods 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 6
- 238000010344 co-firing Methods 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 5
- 230000001105 regulatory effect Effects 0.000 claims description 4
- 238000004891 communication Methods 0.000 claims description 3
- 238000011156 evaluation Methods 0.000 claims description 3
- 238000003062 neural network model Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 3
- 229920006395 saturated elastomer Polymers 0.000 claims description 2
- 230000011664 signaling Effects 0.000 claims description 2
- 238000002474 experimental method Methods 0.000 abstract description 2
- MWUXSHHQAYIFBG-UHFFFAOYSA-N Nitric oxide Chemical compound O=[N] MWUXSHHQAYIFBG-UHFFFAOYSA-N 0.000 description 39
- 230000006870 function Effects 0.000 description 15
- 229910052815 sulfur oxide Inorganic materials 0.000 description 7
- 230000009471 action Effects 0.000 description 6
- 229910000041 hydrogen chloride Inorganic materials 0.000 description 6
- 238000013473 artificial intelligence Methods 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 239000007789 gas Substances 0.000 description 3
- 238000007637 random forest analysis Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 2
- 230000000116 mitigating effect Effects 0.000 description 2
- 230000000306 recurrent effect Effects 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 239000012717 electrostatic precipitator Substances 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 238000003912 environmental pollution Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 239000002803 fossil fuel Substances 0.000 description 1
- 230000002068 genetic effect Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 230000015654 memory Effects 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 230000001699 photocatalysis Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012552 review Methods 0.000 description 1
- XTQHKBHJIVJGKJ-UHFFFAOYSA-N sulfur monoxide Chemical class S=O XTQHKBHJIVJGKJ-UHFFFAOYSA-N 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- -1 uses CO Substances 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
- 239000012719 wet electrostatic precipitator Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
- G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
- G05B13/042—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators in which a parameter or coefficient is automatically adjusted to optimise the performance
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Feedback Control In General (AREA)
- Regulation And Control Of Combustion (AREA)
Abstract
The invention discloses an automatic control method for reducing emission of multi-source biomass flue gas during mixed combustion of a coal-fired boiler, which comprises the following steps: acquiring historical data and real-time data of a target pulverized coal biomass blending furnace, processing and classifying the data, and establishing a database; establishing a DNN-LSTM-Catboost fusion prediction model by using a database, extracting characteristics of the data, and converting the state data into the input of an intelligent agent; training input data by using a DDPG model to obtain an optimal control strategy; and converting the optimal control strategy into an actual control signal according to a decision strategy obtained by the DDPG model, and controlling the incinerator in real time through a hardware real-time controller. Compared with the traditional method for predicting the emission concentration of the smoke pollutants by using experiments and then adjusting the emission concentration, the method provided by the invention is more convenient to drive by using pure data, has high economical efficiency, and can be effectively used for controlling the emission of the pollutants.
Description
Technical Field
The invention relates to the field of pollutant control of pulverized coal biomass co-firing furnaces, in particular to an automatic control method for reducing emission of smoke of multi-source biomass co-firing in a coal-fired boiler.
Background
Biomass resources, which are the third largest world-accepted resource, have become one of the solutions facing the depletion of high-quality coal resources. Worldwide energy consumption has been rapidly increasing, with the development of industrialization, coal is over-mined as an important part of fossil fuels, exhaustion of coal reserves and high-temperature gas emissions thereof have forced people to find an alternative, renewable, less-polluting and environmentally friendly energy source such as biomass. In order to effectively utilize low-quality coal and biomass energy, hybrid combustion has become a research hot spot in recent years.
Due to the difference of fuel composition elements, pollutants discharged by coal in the combustion process are mainly sulfur oxides, CO and particulate matters; the biomass fuel mainly uses CO, NOx and particles, and the pollutants pose a great threat to the safety of the ecological environment, and meanwhile, the risk of damage inside the boiler is increased. In the pulverized coal biomass blending furnace, the blending proportion, the biomass type, the air volume ratio, the excess air coefficient, the oxygen concentration, the air inlet flow rate and the like are all factors influencing the emission characteristics of the smoke pollutants of the furnace.
Methods and techniques for mitigating emissions from flue gas from combustion power plants review conventional techniques for reducing pollutants from exhaust gas from combustion power plants, summarizing the latest methods for reducing pollutants from exhaust gas from combustion power plants, and the prior art can utilize hardware devices (including combined methods of wet electrostatic precipitators and cyclone electrostatic precipitators, microwave systems, photocatalytic techniques, electrochemical systems, etc.) to achieve effective simultaneous removal of multiple pollutants. Artificial intelligence for controlling and optimizing boiler performance and emissions has investigated the use of expert systems, artificial Neural Networks (ANNs), genetic Algorithms (GA), fuzzy Logic (FL) and different hybrid systems in combustion systems to predict and treat boiler performance and emissions. The defects are that: the prior art has mostly focused on the treatment of the generated flue gas with hardware facilities; the technology in the aspect of artificial intelligence adopts a single technology, and the problem of boiler combustion emission reduction is still challenging to treat. The invention uses the control of the combustion process and the combustion process as the starting point of the treatment of the flue gas emission, and combines a plurality of technical algorithms of artificial intelligence, thereby realizing accurate and effective combustion control in real time.
Disclosure of Invention
In order to solve the problems, the invention aims to provide an automatic control method for reducing emission of smoke generated by burning multi-source biomass in a coal-fired boiler.
The invention is realized at least by one of the following technical schemes.
An automatic control method for reducing emission of mixed-combustion multi-source biomass smoke of a coal-fired boiler comprises the following steps:
acquiring historical data and real-time data of a target pulverized coal biomass blending furnace, processing and classifying the data, and establishing a database;
establishing a DNN-LSTM-Catboost fusion prediction model by using a database, extracting characteristics of the data, and converting the state data into the input of an intelligent agent;
training input data by using a DDPG model to obtain an optimal control strategy;
and converting the optimal control strategy into an actual control signal according to a decision strategy obtained by the DDPG model, and controlling the incinerator in real time through a hardware real-time controller.
Further, the trained DNN-LSTM-Catboost fusion prediction model is used for predicting the smoke emission concentration of the co-firing furnace;
the DDPG model is used for giving an optimization strategy for adjusting the parameters of the combustion conditions under the current working condition according to the prediction result of the prediction model and the real-time data;
according to the decision strategy obtained by the DDPG model, converting the optimal control strategy into an actual control signal, and controlling the incinerator in real time through a hardware real-time controller, wherein the method specifically comprises the following steps of:
if the combustion condition parameter optimization strategy obtained by the DDPG model is different from the current parameter, the controller is used for adjusting the combustion condition parameter;
and continuously inputting the parameters regulated by the controller and the regulated real-time data into the DDPG model, thereby obtaining a new optimization strategy in real time.
Further, the DNN-LSTM-Catboost fusion prediction model comprises a DNN-LSTM neural network and a Catboost algorithm;
the DNN-LSTM neural network comprises a parallel network composed of DNN and LSTM and a network with a DNN structure;
and taking the output characteristics of the trained neural network as new characteristics, and inputting the new characteristics into a Catboost algorithm for training, thereby obtaining a prediction model.
Further, the neural network DNN includes a Batchnorm1d layer, a linear layer r, an ELU layer, and a Dropout layer.
Further, the Linear layer sets a hidden layer or a fully connected layer in the network, each neuron in the layer stores the weight of each input feature and the paranoid of the layer, the weight and the paranoid are updated after back propagation calculation, and the Linear layer calculates:
y＝xA T +b
wherein y is the output of each neuron, representing the output result matrix of a batch of data through the Linear layer; the output of each neuron is equal to the input characteristic x, and the corresponding weight A T The sum of the products is added to the offset b.
Further, the ELU layer is an activation function, and according to the linear relation between 5 input parameters of the fuel blending proportion, the air volume ratio, the excess air coefficient, the oxygen concentration and the air inlet flow rate and three output parameters of the HCl, NOx, SOx emission in the smoke, the activation function is introduced to enable the network to have nonlinear expression capacity, and the ELU function expression is as follows:
where x represents the input feature, α represents an adjustable parameter, and ELU is a non-saturated function.
Further, the network of the DNN structure combines a plurality of input features into a parallel network formed by DNN and LSTM and then outputs the features to continue training, a prediction result is obtained, in the previous linear layer of the output result, the input features input into the linear layer are extracted and added into the Catboost algorithm as new features to perform training and evaluation, and therefore a trained DNN-LSTM-Catboost fusion prediction model is obtained, and the subsequent working conditions can be predicted by using the trained fusion prediction model.
Further, when the DNN-LSTM-Catboost fusion prediction model is used for training data, input parameters are working condition data related to the history of the target incinerator and the emission of the smoke pollutants in real time, wherein the working condition data comprises the following components: the fuel blending proportion, the air volume ratio, the excess air coefficient, the oxygen concentration and the air inlet flow rate are input parameters of a model; and selecting the emission amount of HCl, NOx, SOx in the flue gas as an output parameter.
Further, the DDPG model comprises an Actor network for estimating state values and a Critic network for estimating value functions, and the two networks are deep neural network models.
Further, the hardware real-time controller comprises a data acquisition module, a control module and a communication module, and is used for acquiring, processing and sending control signals.
Compared with the prior art, the invention has the beneficial effects that:
in order to reduce the emission of smoke pollutants, the invention predicts the data, controls according to the prediction data, improves the prediction precision, and particularly establishes a DNN-LSTM-Catboost fusion prediction model by utilizing a database, so as to extract the characteristics of the data, and converts the state data into the input of an intelligent agent; training input data by using a DDPG model to obtain an optimal control strategy; and converting the optimal control strategy into an actual control signal according to a decision strategy obtained by the DDPG model, and controlling the incinerator in real time through a hardware real-time controller.
Compared with DNN and DNN-LSTM of the neural network, the Catboost has better performance, and the DNN-LSTM has an LSTM structure, so that time sequence characteristics can be processed more flexibly, and compared with DNN, the DNN-LSTM has higher prediction precision. The fusion model formed by DNN-LSTM-Catoost can enable the Catoost algorithm to learn time sequence information which is difficult to learn, and can fuse the advantages of high Catoost prediction precision and LSTM processing time sequence, so that the prediction precision is further improved. In other embodiments, the fusion model has a 41% reduction in MAE on an as-is basis.
Compared with the traditional method for predicting the emission concentration of the smoke pollutants by using experiments and then adjusting the emission concentration, the method provided by the invention is more convenient to drive by using pure data, has high economical efficiency, and can be effectively used for controlling the emission of the pollutants.
Drawings
FIG. 1 is a flow chart of the automatic control method of the mixing boiler in embodiment 1 of the present invention;
FIG. 2 is a schematic illustration of the design and training of a neural network model of the present invention;
FIG. 3 is a schematic diagram of the agent decision strategy design and training of the present invention.
Detailed Description
The present invention is described in further detail below in connection with specific embodiments, but embodiments of the invention include, but are not limited to, the following examples, which are merely illustrative of the principles of the invention and not in limitation of the practice of the invention. In embodiments, elements of the invention may be substituted, omitted, or altered to achieve the same objectives.
As shown in fig. 1, the embodiment discloses an automatic control method for reducing emission of multi-source biomass flue gas during co-firing of a coal-fired boiler, which comprises the following specific steps:
acquiring historical data and real-time data of a target pulverized coal biomass blending furnace, processing the data, dividing the data into a training set, a verification set and a test set, and establishing a database; acquiring historical data of a target pulverized coal biomass incinerator, and acquiring various parameters in the combustion process by using a sensor to acquire real-time data; processing and classifying the data, and selecting effective data samples to construct a total database; the processing includes cleaning, denoising, normalizing and the like on the data.
Establishing a DNN-LSTM-Catboost fusion prediction model by using a database, extracting characteristics of the data, and converting state data, namely the emission of HCl, NOx, SOx in the flue gas of the output parameter of the prediction model, into the input of an intelligent agent;
the DNN-LSTM-Catboost fusion prediction model comprises a DNN-LSTM neural network structure and a Catboost algorithm;
the DNN-LSTM (long-short-term memory recurrent neural network-deep neural network) neural network comprises a parallel network consisting of DNN and LSTM (long-term memory recurrent neural network) and a network with a DNN structure;
the DNN (deep neural network) is formed by building four structures, namely BatchNorm1d, linear, ELU and Dropout;
the historical data were all subjected to BatchNorm treatment before entering the Linear layer, and equations (1) - (4) describe its principle:
wherein m represents the data amount, x i The representation represents the ith data; equation (1) is an average value of one lot, and equation (2) is a variance equation. Mu (mu) β Mean value of batch data,Variance, th ∈of one lot of data>Representing the normalized result matrix, y, of the data in a batch of data i The data representing a batch passes through the output result matrix of the BatchNorm1d layer.
As an example, for example, there are 10 data in total, then m is 10, x 1 、x 2 Up to x 10 Namely, 10 specific data are summed and then divided to obtain an average value;
equation (3) normalizes (normalizes) the data for a batch for a feature, e is a small number to ensure that the denominator is not zero. Equation (1) and equation (2) are the mean and variance, respectively, of the batch data. In the formula (4), gamma and beta are two trainable parameters, the initial values are 1 and 0, and the parameters are updated along with back propagation;
the Linear sets a hidden layer or a fully connected layer in the network, each neuron in the layer stores the weight of each input feature and the paranoid of the layer, and the weight and the paranoid are updated after back propagation calculation (the paranoid can be not updated). Equation (5) describes the calculation process of the Linear layer:
y＝xA T +b (5)
y represents the output result matrix of a batch of data through the Linear layer. The output of each neuron is equal to the input characteristic x, and the corresponding weight A T The sum of the products is added to the offset b.
As an example, if a batch of data is 64, there are 5 input parameters, the input data format is 64×5, if the first layer Linear has 256 neurons, the weight matrix format is 256×5, the transposed result is 5×256, and the output format after multiplication is 64×256. The 5 features corresponding to each group of data are converted into 256 features after passing through the Linear layer;
ELU is one type of activation function. Considering that the three output parameters of 5 input parameters of fuel blending proportion, air volume ratio, excess air coefficient, oxygen concentration and air inlet flow rate and the emission of HCl, NOx, SOx in the flue gas are not in a common linear relation, an activation function needs to be introduced to enable the network to have nonlinear expression capability. Equation (6) is an ELU function expression:
where α represents an adjustable parameter, whose value is greater than 0, and the default value is 1.ELU is an unsaturated function, and does not suffer from gradient explosion or disappearance;
introducing Dropout layers can be relatively effective in mitigating overfitting, which can cause neurons at a layer in the network to be inactivated (temporarily zeroed) with a certain probability.
As an embodiment, if the probability of the Dropout layer is set to 0.2, then the probability of 0.2 for each feature input to the Dropout layer is temporarily zeroed out;
wherein, the LSTM related patent and literature are not described in detail herein;
the historical data and the real-time data are selected and respectively input into DNN and LSTM networks for training, and the output characteristics of the historical data and the real-time data are spliced and then are sent into DNN networks with similar structures as the first layer for training;
the other layer is a network with a DNN structure, a plurality of input features enter a parallel network formed by DNN and LSTM and then output features are combined to continue training, a prediction result is obtained, in the previous linear layer of the output result, the input features input into the linear layer are extracted and added into Catboost as new features to perform training and evaluation, and therefore a trained DNN-LSTM-Catboost fusion prediction model is obtained, and the subsequent working conditions can be predicted by using the trained fusion prediction model.
When the DNN-LSTM-Catboost fusion prediction model is used for training data, input parameters are target incinerator history and real-time working condition data related to the emission of smoke pollutants, wherein the DNN-LSTM-Catboost fusion prediction model comprises the following components: the fuel blending proportion, the air volume ratio, the excess air coefficient, the oxygen concentration and the air inlet flow rate are input parameters of a model; selecting the emission amount of HCl, NOx, SOx in the flue gas as an output parameter;
the Catboost algorithm, currently common integration algorithms, are two: bagging and Boosting. The random forest algorithm is one of Bagging algorithms. XGBoost, lightGBM, catBoost is an improved implementation of the Gradient Boost Decision Tree (GBDT), all of which belong to the Boosting algorithm. In the fields of environment and energy, the algorithms are not used for solving the prediction problem, and the effect is not popular;
often perform better in prediction accuracy than XGBoost and LihgtGBM, catBoost. After training and predicting the same dataset by using random forests, XGBoost, lightGBM, catBoost, the results showed that Catboost was more accurate than XGBoost, lightGBM. Compared with a random forest based on Bagging, the method is more stable in prediction.
The trained DNN-LSTM-Catboost fusion prediction model is used for predicting the emission concentration of the smoke pollutants of the co-firing furnace.
Training input data by using a depth deterministic strategy gradient algorithm (DDPG) model to obtain an optimal control strategy; the DDPG model is used for giving an optimization strategy for adjusting the parameters of the combustion conditions under the current working condition according to the prediction result of the prediction model and the real-time data;
the intelligent agent is used for realizing the online real-time control of the combustion process.
In decision strategy design and training of an intelligent agent, firstly, a state space and an action space of a combustion process are divided, wherein the state space comprises state quantities such as HCl, NOx, SOx emission in flue gas, and the action space comprises control variables such as fuel blending proportion, air volume ratio, excess air coefficient, oxygen concentration, air inlet flow rate and the like;
a reward function is defined for evaluating feedback obtained by an agent taking an action in a certain state. The reward function should be able to encourage the agent to take the correct action to achieve effective control of the incinerator control system.
Wherein the reward function for NOx:
wherein C is NOx Is the nitrogen oxide (NOx) concentration T per moment is the length of the time window, R t Is the prize value at time t. The implication of this reward function is to minimize the average of the square of NOx over a period of time. This may encourage the agent to choose control strategies that keep the NOx concentration as low as possible, thereby reducing environmental pollution.
The strategy of the intelligent agent is trained by adopting a depth deterministic strategy gradient algorithm (DDPG model) so as to obtain an optimal control strategy.
The decision model of the intelligent agent adopts an Actor-Critic model, and comprises a strategy network Actor neural network and a comment network Critic neural network. The Actor neural network is used for outputting decision strategies of the intelligent agent, and the Critic neural network is used for evaluating the quality of the decision strategies output by the Actor network. An agent refers to an entity trained using the DDPG algorithm with the goal of optimizing fuel ratios to achieve the goal of reducing pollutant emissions. The policies of the agent are represented by an Actor network, and a Critic network is used to evaluate the performance of the policies.
The process of training an agent can be divided into two phases: training of an Actor network and training of an Actor-Critic network. In training an Actor network, DDPG is used to train the Actor neural network so that the Actor neural network gradually learns the optimal fuel ratio strategy.
In the training of the Actor-Critic network, the output of the Actor neural network is used as an action, the next state is obtained through an environment transfer function, then the state and the action are input into the Critic neural network for value estimation, and finally the network parameters are updated according to a reward function so as to optimize the decision strategy of the intelligent agent.
Through continuous iterative training, the Actor neural network gradually learns the optimal fuel proportioning strategy.
And converting the optimal control strategy into an actual control signal according to a decision strategy obtained by the DDPG model, and controlling the incinerator in real time through a hardware real-time controller.
The hardware real-time controller comprises a data acquisition module, a control module and a communication module, and is used for acquiring, processing and sending control signals.
The hardware real-time controller selects a PLC controller, converts an optimal control strategy output by a DDPG algorithm into an actual control signal, and realizes real-time control of the mixed combustion furnace by controlling parameters such as fuel mixing proportion, air volume ratio, excess air coefficient, oxygen concentration, air inlet flow rate and the like.
The preferred embodiments of the invention disclosed above are intended only to assist in the explanation of the invention. The preferred embodiments are not exhaustive or to limit the invention to the precise form disclosed. Obviously, many modifications and variations are possible in light of the above teaching. The embodiments were chosen and described in order to best explain the principles of the invention and the practical application, to thereby enable others skilled in the art to best understand and utilize the invention.
Claims (10)
1. An automatic control method for reducing emission of mixed-combustion multi-source biomass smoke of a coal-fired boiler is characterized by comprising the following steps of: the method comprises the following steps:
acquiring historical data and real-time data of a target pulverized coal biomass blending furnace, processing and classifying the data, and establishing a database;
establishing a DNN-LSTM-Catboost fusion prediction model by using a database, extracting characteristics of the data, and converting the state data into the input of an intelligent agent;
training input data by using a DDPG model to obtain an optimal control strategy;
and converting the optimal control strategy into an actual control signal according to a decision strategy obtained by the DDPG model, and controlling the incinerator in real time through a hardware real-time controller.
2. The automatic control method for reducing emission of co-fired multi-source biomass flue gas of the coal-fired boiler according to claim 1, which is characterized by comprising the following steps: the trained DNN-LSTM-Catboost fusion prediction model is used for predicting the smoke emission concentration of the co-firing furnace;
the DDPG model is used for giving an optimization strategy for adjusting the parameters of the combustion conditions under the current working condition according to the prediction result of the prediction model and the real-time data;
according to the decision strategy obtained by the DDPG model, converting the optimal control strategy into an actual control signal, and controlling the incinerator in real time through a hardware real-time controller, wherein the method specifically comprises the following steps of:
if the combustion condition parameter optimization strategy obtained by the DDPG model is different from the current parameter, the controller is used for adjusting the combustion condition parameter;
and continuously inputting the parameters regulated by the controller and the regulated real-time data into the DDPG model, thereby obtaining a new optimization strategy in real time.
3. The automatic control method for reducing emission of the mixed combustion multi-source biomass flue gas of the coal-fired boiler according to claim 1, wherein the DNN-LSTM-Catboost fusion prediction model comprises a DNN-LSTM neural network and a Catboost algorithm;
the DNN-LSTM neural network comprises a parallel network composed of DNN and LSTM and a network with a DNN structure;
and taking the output characteristics of the trained neural network as new characteristics, and inputting the new characteristics into a Catboost algorithm for training, thereby obtaining a prediction model.
4. The automatic control method for reducing emission of co-fired multi-source biomass flue gas of a coal-fired boiler according to claim 3, wherein the neural network DNN comprises a BatchNorm1d layer, a linear layer r, an ELU layer and a Dropout layer.
5. The automatic control method for reducing emission of co-fired multi-source biomass flue gas of a coal-fired boiler according to claim 4, wherein a hidden layer or a full-connection layer in a network is arranged on a Linear layer, each neuron in the layer stores weight of each input characteristic and the paranoid of the layer, the weight and the paranoid are updated after counter-propagation calculation, and the calculation process of the Linear layer comprises the following steps of:
y＝xA T +b
wherein y is the output of each neuron, representing the output result matrix of a batch of data through the Linear layer; the output of each neuron is equal to the input characteristic x, and the corresponding weight A T The sum of the products is added to the offset b.
6. The automatic control method for reducing emission of multi-source biomass flue gas for coal-fired boiler according to claim 3, wherein the ELU layer is an activation function, the activation function is introduced to enable the network to have nonlinear expression capability according to the linear relation between 5 input parameters of fuel blending proportion, air volume ratio, excess air coefficient, oxygen concentration and air inlet flow rate and the output parameters of HCl, NOx, SOx in the flue gas, and the ELU function expression is as follows:
where x represents the input feature, α represents an adjustable parameter, and ELU is a non-saturated function.
7. The automatic control method for reducing emission of the mixed-burning multi-source biomass flue gas of the coal-fired boiler according to claim 3, wherein the network of the DNN structure combines a plurality of input features into a parallel network formed by DNN and LSTM and then outputs the features to continue training, a prediction result is obtained, the input features input into the linear layer are extracted from the previous linear layer of the output result and are added into a Catboost algorithm as new features to perform training and evaluation, so that a trained DNN-LSTM-Catboost fusion prediction model is obtained, and the subsequent working conditions can be predicted by using the trained fusion prediction model.
8. The automatic control method for reducing emission of co-fired multi-source biomass flue gas of a coal-fired boiler according to claim 1, wherein when the DNN-LSTM-Catboost fusion prediction model is used for training data, input parameters are target incinerator history and real-time working condition data related to flue gas pollutant emission, and the method comprises the following steps: the fuel blending proportion, the air volume ratio, the excess air coefficient, the oxygen concentration and the air inlet flow rate are input parameters of a model; and selecting the emission amount of HCl, NOx, SOx in the flue gas as an output parameter.
9. The automatic control method for reducing emission of co-fired multi-source biomass flue gas of a coal-fired boiler according to claim 1, wherein the DDPG model comprises an Actor network for state value estimation and a Critic network for value function estimation, and the two networks are deep neural network models.
10. The automatic control method for reducing emission of the co-fired multi-source biomass flue gas of the coal-fired boiler according to claim 1, wherein the hardware real-time controller comprises a data acquisition module, a control module and a communication module and is used for acquiring, processing and sending control signals.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311263477.4A CN117471906A (en) | 2023-09-27 | 2023-09-27 | Automatic control method for reducing emission of mixed-combustion multi-source biomass smoke of coal-fired boiler |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311263477.4A CN117471906A (en) | 2023-09-27 | 2023-09-27 | Automatic control method for reducing emission of mixed-combustion multi-source biomass smoke of coal-fired boiler |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117471906A true CN117471906A (en) | 2024-01-30 |
Family
ID=89630192
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311263477.4A Pending CN117471906A (en) | 2023-09-27 | 2023-09-27 | Automatic control method for reducing emission of mixed-combustion multi-source biomass smoke of coal-fired boiler |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117471906A (en) |
-
2023
- 2023-09-27 CN CN202311263477.4A patent/CN117471906A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Xia et al. | Dioxin emission prediction based on improved deep forest regression for municipal solid waste incineration process | |
| CN103576655B (en) | A kind of power boiler burning subspace modeling and Multipurpose Optimal Method and system | |
| CN111144609A (en) | Boiler exhaust emission prediction model establishing method, prediction method and device | |
| CN113742997B (en) | Intelligent optimization setting method for air quantity in urban solid waste incineration process | |
| CN112488145A (en) | NO based on intelligent methodxOnline prediction method and system | |
| CN112613237B (en) | CFB unit NOx emission concentration prediction method based on LSTM | |
| CN111589302A (en) | Method, device, equipment and storage medium for predicting SCR denitration performance of coal-fired power plant | |
| CN110991756A (en) | MSWI furnace temperature prediction method based on TS fuzzy neural network | |
| CN114398836A (en) | MSWI process dioxin emission soft measurement method based on width mixed forest regression | |
| CN104696952A (en) | Fluidized bed boiler air input capacity control method based on model online identification | |
| Manjunatha et al. | Application of Artificial Neural Networks for emission modelling of biodiesels for a CI engine under varying operating conditions | |
| Qiao | A novel online modeling for NOx generation prediction in coal-fired boiler | |
| Xu et al. | A novel online combustion optimization method for boiler combining dynamic modeling, multi-objective optimization and improved case-based reasoning | |
| CN110888401B (en) | Combustion control optimization method and device for thermal generator set and readable storage medium | |
| CN113947013A (en) | Boiler short-term NO based on hybrid deep neural network modelingxEmission prediction method | |
| WO2023231667A1 (en) | Method for soft measurement of dioxin emission in mswi process based on integrated t-s fuzzy regression tree | |
| CN117471906A (en) | Automatic control method for reducing emission of mixed-combustion multi-source biomass smoke of coal-fired boiler | |
| Ilamathi et al. | Predictive modelling and optimization of nitrogen oxides emission in coal power plant using Artificial Neural Network and Simulated Annealing | |
| Huang et al. | Large-Scale and Knowledge-Based Dynamic Multiobjective Optimization for MSWI Process Using Adaptive Competitive Swarm Optimization | |
| Balamurugan et al. | ANN-SQP Approach for NOx emission reduction in coal fired Boilers | |
| Yan et al. | Model Prediction and Optimal Control of Gas Oxygen Content for A Municipal Solid Waste Incineration Process | |
| Ilamathi et al. | A novel approach for modelling of NOx emission reduction in a tangentially fired coal boiler | |
| CN115542740B (en) | Multivariable active disturbance rejection control method based on data driving for urban solid waste incineration process | |
| CN113847609B (en) | Garbage incinerator denitration control method based on feedforward model prediction error self-correction | |
| CN115717709B (en) | Method for predicting heat value of garbage in furnace in real time based on attention mechanism LSTM model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |