Coal-fired boiler combustion optimization method, system, device and medium coupled with water wall temperature constraints

Abstract

This invention discloses a method, system, equipment, and medium for optimizing the combustion of coal-fired boilers with coupled water-cooled wall temperature constraints. It relates to the field of intelligent operation optimization and safety control technology for large-scale coal-fired boilers, and solves the technical problems of insufficient modeling accuracy of boiler combustion processes, single optimization objectives, and lack of water-cooled wall safety constraints in existing technologies. The key technical solution is to integrate boiler efficiency, NOx emission concentration, regional temperature difference of the water-cooled wall, and the highest regional wall temperature into the same optimization framework, and to set safety boundary constraints and action-holding constraints. Historical operating samples are divided into operating conditions, and combustion state prediction models are established on different subsets of operating conditions, reducing the adverse impact of data distribution differences in different operating intervals on the modeling results. Ultimately, this enables more accurate prediction of key indicators such as boiler efficiency, NOx emission concentration, regional temperature difference of the water-cooled wall, and the highest regional wall temperature under corresponding operating conditions.

Description

Technical Field

[0001] This application relates to the field of intelligent operation optimization and safety control technology for large coal-fired boilers, and in particular to a method, system, equipment and medium for combustion optimization of coal-fired boilers coupled with water-cooled wall temperature constraints. Background Technology

[0002] With the increasing proportion of new energy installed capacity, coal-fired power units in the power system need to undertake more frequent and larger-scale load regulation tasks. Under deep peak shaving and frequent load change operation conditions, the boiler combustion process is jointly affected by changes in coal distribution, air distribution mode, and in-furnace flow and heat transfer state, exhibiting obvious nonlinearity, strong coupling of multiple variables, and mutual constraints in operation. On the one hand, load fluctuations and combustion adjustments can easily lead to changes in the formation mechanism of nitrogen oxides (NOx), making it difficult to stably control emission levels; on the other hand, during low-load or frequent load change operation, boiler efficiency often shows a downward trend, affecting the overall economic efficiency of the unit. At the same time, during the deep peak shaving operation stage, due to the offset of the combustion center in the furnace, uneven heat load distribution, and reduced hydrodynamic circulation capacity of the heating surface, local wall temperature rises or regional wall temperature differences are easily caused in the water-cooled wall, thereby increasing the risk of thermal fatigue, corrosion, and tube rupture of the heating surface, affecting the safe and stable operation of the boiler.

[0003] Existing boiler combustion optimization methods mostly rely on manual experience or single-objective control and measurement, making it difficult to achieve synergistic optimization of pollutant emissions, combustion efficiency and water-cooled wall safety under complex and variable operating conditions. This makes it difficult to meet the comprehensive operational requirements of current coal-fired units in terms of environmental protection, economy and safety. Summary of the Invention

[0004] This application provides a method, system, equipment, and medium for optimizing the combustion of a coal-fired boiler with coupled water-cooled wall temperature constraints. The technical objective is to effectively suppress local overheating and thermal deviation in the water-cooled wall, while improving the modeling accuracy of the boiler combustion process, enriching the optimization objectives, and meeting the comprehensive operational needs of coal-fired units in terms of environmental protection, economy, and safety.

[0005] The above-mentioned technical objective of this application is achieved through the following technical solution: A combustion optimization method for a coal-fired boiler coupled with water-cooled wall temperature constraints includes: Step 1: Acquire historical operating data of the coal-fired boiler under different operating conditions; wherein, the historical operating data includes operating parameters and performance parameters, the operating parameters include boiler load, coal feed rate of each coal mill, total air supply, each air supply branch, opening degree of secondary air damper of each layer and oxygen content in flue gas, and the performance parameters include boiler efficiency, NOx emission concentration and water-cooled wall measuring point temperature; Step 2: Preprocess the historical running data to remove samples containing invalid data and obtain a continuous multivariate time series sample dataset; Step 3: Select boiler load, total coal feed and total air feed as operating condition discriminant variables from the historical operating data, perform cluster analysis on the feature vector formed by the operating condition discriminant variables, and divide the historical operating data into 3 typical operating conditions based on the results of the cluster analysis. Step 4: Using boiler efficiency, NOx emission concentration, regional temperature difference of water-cooled wall and regional maximum wall temperature as output targets, select corresponding input auxiliary variables from the multivariate time series sample dataset according to the output targets, perform feature filtering on the input auxiliary variables, and obtain the final input feature set; Step 5: Construct and train a combustion state prediction model for each typical operating condition; wherein, the combustion state prediction model is constructed using a combination of convolutional neural network and bidirectional long short-term memory network. Step 6: With the optimization objectives of reducing NOx emission concentration, improving boiler efficiency, reducing temperature difference in each water-cooled wall region, and reducing the maximum wall temperature of the water-cooled wall, construct the comprehensive objective function of the combustion state prediction model; Step 7: Within the safety boundary constraints, the control variable space is searched using a multi-objective intelligent optimization algorithm. The control variables are iteratively updated using the combustion state prediction model to minimize the comprehensive objective function, thereby obtaining the optimal control variable vector. The boiler combustion is then optimized and adjusted based on the optimal control variable vector. The safety boundary constraints are the minimum and maximum value boundaries corresponding to each parameter in the historical operating data. The control variables include the opening degree of each secondary damper, the total air supply, and the coal feed rate of each coal mill.

[0006] Preferably, the cluster analysis employs the fuzzy C-means clustering method, and its objective function is expressed as: ; in, Indicates the first Feature vectors corresponding to historical operational data This represents the total number of samples in the historical operational data; Indicates the first Cluster centers corresponding to each operating condition category Indicates the number of operating condition categories; Representing the eigenvector For the Membership degree of each operating condition category; This represents the fuzziness index.

[0007] Preferably, in step 4, feature filtering is performed on the input auxiliary variables to obtain the final input feature set, including: The input auxiliary variables are feature-selected using a random forest regression model. The random forest regression model calculates the importance index of the input auxiliary variables using out-of-bag error perturbation, as follows: ; in, Indicates the first Importance indicators of input auxiliary variables Indicates the number of decision trees. and These represent the out-of-bag data before and after the perturbation at the [number]th ... Prediction error on each decision tree; The importance indicators are sorted in descending order, and the top K importance indicators are selected. The input auxiliary variables corresponding to these top K importance indicators are the important features, and the union of these important features is the final set of input features.

[0008] Preferably, the input sequence of the combustion state prediction model is: a continuous sequence of features in the final input feature set. If a time series is composed of characteristic variables at historical moments, then the time series is represented as: ; in, express The final input at any given time corresponds to the feature variable in the feature set. The output of the combustion state prediction model is expressed as follows: ; in and They represent The first water-cooled wall Temperature difference in each region and highest wall temperature in each region, express Boiler efficiency at any given time express NOx emission concentration at any given time.

[0009] Preferably, the loss function of the combustion state prediction model is expressed as: ; in, Indicates the number of training samples. Represents the true value. This represents the predicted value output by the combustion state prediction model.

[0010] Preferably, the comprehensive objective function is expressed as: ; ; in, Represents the overall objective function. This indicates that the action remains constrained. , , and These represent the changes in NOx emission concentration, boiler efficiency, temperature difference in each water-cooled wall region, and maximum water-cooled wall temperature relative to the current operating state under the action of candidate control actions. , , and These represent the statistical scales for NOx emission concentration, boiler efficiency, temperature difference in each water-cooled wall region, and maximum water-cooled wall temperature, respectively. , , , , All represent weighting coefficients; Indicates the first Candidate values ​​for each control variable; This indicates the value of the corresponding control variable under the current operating condition. This indicates the allowable range of variation for the control variable. This indicates the total number of control variables.

[0011] Preferably, in step 2, the preprocessing method includes outlier removal and missing value processing.

[0012] A combustion optimization system for a coal-fired boiler coupled with water-cooled wall temperature constraints, the system being used to implement the aforementioned coal-fired boiler combustion optimization method, the system comprising: The data acquisition module acquires historical operating data of the coal-fired boiler under different operating conditions. The historical operating data includes operating parameters and performance parameters. The operating parameters include boiler load, coal feed rate of each coal mill, total air supply, air flow branches, secondary air damper opening of each layer, and flue gas oxygen content. The performance parameters include boiler efficiency, NOx emission concentration, and water-cooled wall measuring point temperature. The preprocessing module preprocesses the historical running data to remove samples containing invalid data, thereby obtaining a continuous multivariate time series sample dataset. The clustering module selects boiler load, total coal feed and total air feed as operating condition discriminant variables from the historical operating data, performs cluster analysis on the feature vector formed by the operating condition discriminant variables, and divides the historical operating data into 3 typical operating conditions based on the results of the cluster analysis. The feature filtering module takes boiler efficiency, NOx emission concentration, temperature difference of each water-cooled wall area and maximum water-cooled wall temperature as output targets, selects corresponding input auxiliary variables from the multivariate time series sample dataset according to the output targets, performs feature filtering on the input auxiliary variables, and obtains the final input feature set. The model building module constructs and trains a combustion state prediction model for each typical operating condition; wherein, the combustion state prediction model is constructed using a combination of convolutional neural network and bidirectional long short-term memory network. The objective construction module constructs the comprehensive objective function of the combustion state prediction model with the optimization objectives of reducing NOx emission concentration, improving boiler efficiency, reducing temperature difference in each water-cooled wall region, and reducing the maximum wall temperature of the water-cooled wall. The target optimization module searches the control variable space within the safety boundary constraints using a multi-objective intelligent optimization algorithm, iteratively updates the control variables using the combustion state prediction model to minimize the comprehensive objective function, thereby obtaining the optimal control variable vector, and then optimizes and adjusts boiler combustion based on the optimal control variable vector.

[0013] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the steps of the coal-fired boiler combustion optimization method.

[0014] A computer storage medium storing a computer program, which, when executed by a processor, implements the steps of the coal-fired boiler combustion optimization method.

[0015] The above technical solution can achieve at least some of the following technical effects: The coal-fired boiler combustion optimization method, system, equipment and medium with coupled water-cooled wall temperature constraints described in this application integrates boiler efficiency, NOx emission concentration, regional temperature difference of water-cooled wall and regional maximum wall temperature into the same optimization framework, and sets safety boundary constraints and action holding constraints, so that the optimization results can not only improve the economic and environmental performance of boiler operation, but also effectively suppress the phenomenon of local overheating and thermal deviation of water-cooled wall, thereby improving the safe and stable operation capability of boiler under deep peak shaving and variable load operation conditions.

[0016] Furthermore, due to significant differences in combustion characteristics, heat transfer processes, and multivariate coupling relationships under different boiler operating conditions, traditional unified modeling methods for all operating conditions struggle to maintain prediction accuracy across various scenarios. This is particularly problematic under complex operating conditions such as low load and frequent load changes, where the model's generalization ability tends to decline. This application addresses this issue by dividing historical operating samples into different operating condition subsets and establishing separate combustion state prediction models for each subset. This reduces the adverse impact of data distribution differences across operating intervals on the modeling results, enabling more accurate predictions of key indicators such as boiler efficiency, NOx emission concentration, regional temperature difference of water-cooled walls, and regional maximum wall temperature under the corresponding operating conditions. This provides a more reliable model foundation for subsequent multi-objective optimization. Attached Figure Description

[0017] Figure 1 This is a flowchart of the combustion optimization method for a coal-fired boiler coupled with water-cooled wall temperature constraints in the embodiments of this application; Figure 2a This is a schematic diagram illustrating the predicted boiler efficiency in an embodiment of this application. Figure 2b This is a schematic diagram illustrating the predicted NOx emission concentration in the embodiments of this application; Figure 2c This is a schematic diagram showing the predicted regional temperature difference of the water-cooled wall in an embodiment of this application. Figure 2d This is a schematic diagram showing the predicted results of the highest regional wall temperature of the water-cooled wall in the embodiments of this application; Figure 3a This is a schematic diagram illustrating the optimized load of the coal-fired boiler section in the embodiments of this application; Figure 3b This is a schematic diagram illustrating the optimized boiler efficiency in an embodiment of this application. Figure 3c This is a schematic diagram illustrating the optimized NOx emission concentration results in the embodiments of this application; Figure 3d This is a schematic diagram showing the optimized regional temperature difference of the water-cooled wall in an embodiment of this application. Figure 3e This is a schematic diagram showing the optimized results of the highest regional wall temperature of the water-cooled wall in the embodiments of this application. Detailed Implementation

[0018] The technical solution of this application will be described in detail below with reference to the accompanying drawings.

[0019] like Figure 1 As shown, the combustion optimization method for coal-fired boilers with coupled water-cooled wall temperature constraints described in this application includes: Step 1: Acquire historical operating data of the coal-fired boiler under different operating conditions; wherein, the historical operating data includes operating parameters and performance parameters, the operating parameters include boiler load, coal feed rate of each coal mill, total air supply, each air supply branch, opening degree of secondary air damper of each layer and oxygen content in flue gas, and the performance parameters include boiler efficiency, NOx emission concentration and water-cooled wall measuring point temperature.

[0020] Step 2: Preprocess the historical running data to remove samples containing invalid data and obtain a continuous multivariate time series sample dataset.

[0021] Preferably, in step 2, the preprocessing method includes outlier removal and missing value processing.

[0022] Step 3: Select boiler load, total coal feed and total air feed as operating condition discriminant variables from the historical operating data, perform cluster analysis on the feature vector formed by the operating condition discriminant variables, and divide the historical operating data into 3 typical operating conditions based on the results of the cluster analysis.

[0023] Preferably, the cluster analysis employs the fuzzy C-means clustering method, and its objective function is expressed as: ; in, Indicates the first Feature vectors corresponding to historical operational data This represents the total number of samples in the historical operational data; Indicates the first Cluster centers corresponding to each operating condition category Indicates the number of operating condition categories; Representing the eigenvector For the Membership degree of each operating condition category; This represents the fuzziness index, used to adjust the degree of fuzziness in the membership distribution.

[0024] Step 4: Using boiler efficiency, NOx emission concentration, regional temperature difference of water-cooled wall, and regional maximum wall temperature as output targets, select corresponding input auxiliary variables from the multivariate time series sample dataset according to the output targets, perform feature filtering on the input auxiliary variables, and obtain the final input feature set.

[0025] Preferably, in step 4, feature filtering is performed on the input auxiliary variables to obtain the final input feature set, including: The input auxiliary variables are feature-selected using a random forest regression model. The random forest regression model calculates the importance index of the input auxiliary variables using out-of-bag error perturbation, as follows: ; in, Indicates the first Importance indicators of input auxiliary variables Indicates the number of decision trees. and These represent the out-of-bag data before and after the perturbation at the [number]th ... Prediction error on each decision tree; The importance indicators are sorted in descending order, and the top K importance indicators are selected. The input auxiliary variables corresponding to these top K importance indicators are the important features, and the union of these important features is the final set of input features.

[0026] Step 5: Construct and train a combustion state prediction model for each typical operating condition; wherein, the combustion state prediction model is constructed using a combination of convolutional neural network and bidirectional long short-term memory network, and achieves joint prediction of multi-target combustion states through a multi-output method.

[0027] Preferably, the combustion state prediction model constructs the input sequence using a sliding time window method, then the input sequence is: the continuous sequence in the final input feature set. If a time series is composed of characteristic variables at historical moments, then the time series is represented as: ; in, express The final input at any given time corresponds to the feature variable in the feature set. The output of the combustion state prediction model is expressed as follows: ; in and They represent The first water-cooled wall Temperature difference in each region and highest wall temperature in each region, express Boiler efficiency at any given time express NOx emission concentration at any given time.

[0028] Specifically, the combustion state prediction model first performs convolution operations on the input sequence along the time dimension through a one-dimensional convolutional layer to extract short-term fluctuation features and local coupling responses during the combustion process; then, a second convolutional layer is used to further enhance the expressive power of local dynamic patterns; and at the end of the network, a fully connected layer is used to realize multi-output regression, which is used to simultaneously output the predicted values ​​of boiler efficiency, NOx emissions, and temperature difference and maximum wall temperature of multiple water-cooled wall regions.

[0029] Under each operating condition, the combustion state prediction model is trained based on the training samples of the corresponding operating condition. The model parameters are iteratively updated by minimizing the error function between the model prediction output and the actual combustion state parameters until the preset convergence condition is met.

[0030] Preferably, the loss function of the combustion state prediction model is expressed as: ; in, Indicates the number of training samples. Represents the true value. This represents the predicted value output by the combustion state prediction model. The model parameters are processed using a backpropagation algorithm. The model is iteratively updated and trained to obtain a combustion state prediction model for different working conditions, which is then used for subsequent prediction and optimization.

[0031] Step 6: With the optimization objectives of reducing NOx emission concentration, improving boiler efficiency, reducing temperature difference in each water-cooled wall region, and reducing the maximum wall temperature of the water-cooled wall, construct the comprehensive objective function of the combustion state prediction model.

[0032] Preferably, the comprehensive objective function is expressed as: ; ; in, Represents the overall objective function. This indicates that the action remains constrained. , , and These represent the changes in NOx emission concentration, boiler efficiency, temperature difference in each water-cooled wall region, and maximum water-cooled wall temperature relative to the current operating state under the action of candidate control actions. , , and These represent the statistical scales for NOx emission concentration, boiler efficiency, temperature difference in each water-cooled wall region, and maximum water-cooled wall temperature, respectively, and are used to perform dimensionless normalization on different physical quantities. , , , , Each represents a weighting coefficient, used to adjust the relative importance of each objective in the overall optimization; Indicates the first Candidate values ​​for several control variables, including the opening degree of secondary air dampers at each level, total air supply, and coal feed rate of each coal mill; This indicates the value of the corresponding control variable under the current operating condition; This indicates the allowable range of variation for the control variable, used to standardize the scale of variation for different control variables; This indicates the total number of control variables.

[0033] Step 7: Within the safety boundary constraints, search the control variable space using a multi-objective intelligent optimization algorithm, and apply the combustion state prediction model to the control variables. Perform iterative updates to minimize the comprehensive objective function. Thus, the optimal control variable vector is obtained. According to the optimal control variable vector The boiler combustion is optimized and adjusted; wherein, the safety boundary constraints are the minimum and maximum value boundaries corresponding to each parameter in the historical operating data.

[0034] Specifically, in each iteration, the predicted combustion state value corresponding to the candidate control action is calculated based on the combustion state prediction model, and the comprehensive objective function value is calculated accordingly. When the optimization algorithm meets the preset convergence condition or reaches the maximum number of iterations, the corresponding optimal control variable vector is output. The optimal control variable is used as the boiler combustion regulation strategy.

[0035] The coal-fired boiler combustion optimization system with coupled water-cooled wall temperature constraints described in this application includes a data acquisition module, a preprocessing module, a clustering module, a feature selection module, a model building module, a target building module, and a target optimization module.

[0036] The data acquisition module is used to acquire historical operating data of coal-fired boilers under different operating conditions. The historical operating data includes operating parameters and performance parameters. The operating parameters include boiler load, coal feed rate of each coal mill, total air supply, air flow branches, secondary air damper opening of each layer, and flue gas oxygen content. The performance parameters include boiler efficiency, NOx emission concentration, and water-cooled wall measuring point temperature.

[0037] The preprocessing module is used to preprocess the historical running data to remove samples containing invalid data and obtain a continuous multivariate time series sample dataset.

[0038] The clustering module is used to select boiler load, total coal feed and total air feed as operating condition discriminant variables from the historical operating data, perform cluster analysis on the feature vector formed by the operating condition discriminant variables, and divide the historical operating data into 3 typical operating conditions based on the results of the cluster analysis.

[0039] The feature filtering module is used to select corresponding input auxiliary variables from the multivariate time series sample dataset based on the output targets, such as boiler efficiency, NOx emission concentration, temperature difference of each water-cooled wall region, and maximum wall temperature of the water-cooled wall, and then perform feature filtering on the input auxiliary variables to obtain the final input feature set.

[0040] The model building module is used to build and train the combustion state prediction model corresponding to each typical operating condition; wherein, the combustion state prediction model is built using a combination of convolutional neural network and bidirectional long short-term memory network.

[0041] The objective construction module is used to construct the comprehensive objective function of the combustion state prediction model with the optimization objectives of reducing NOx emission concentration, improving boiler efficiency, reducing temperature difference in each water-cooled wall region, and reducing the maximum wall temperature of the water-cooled wall.

[0042] The objective optimization module is used to search the control variable space within the safety boundary constraints using a multi-objective intelligent optimization algorithm. It iteratively updates the control variables using the combustion state prediction model to minimize the comprehensive objective function, thereby obtaining the optimal control variable vector. The boiler combustion is then optimized and adjusted based on the optimal control variable vector.

[0043] This application constructs an integrated combustion state prediction model for different operating conditions, considering boiler efficiency, NOx emissions, and water-cooled wall temperature safety indicators, under the constraints of boiler safety boundary operation. This combustion state prediction model is coupled with a multi-objective optimization algorithm to form a practical combustion optimization decision-making framework. The following is combined with… Figure 1 The embodiments of this application are described in detail with reference to Table 1: Step 1: This embodiment uses a 1000MW ultra-supercritical coal-fired unit as the research object. The unit is a double-tangential boiler. Historical operating data of DCS was collected, with a sampling interval of 6 months and a sampling period of 1 minute, for a total of 286,562 samples. Under each operating condition, 70% was selected as the training set, 15% as the test set, and 15% as the validation set. The historical data includes: unit load, total coal feed and coal feed of each coal mill, total air feed and air flow branches, secondary damper opening of each floor, oxygen content in flue gas, and other operating variables, as well as boiler efficiency, SCR inlet NOx emission concentration, water-cooled wall measuring point temperature, and other performance variables.

[0044] Step 2: In this embodiment, the temperature data from the water-cooled wall measuring points are first processed. Specifically, based on the spatial structure of the boiler water-cooled wall, the front wall, rear wall, left wall, and right wall are divided into several regions. For each moment... Temperature at all measuring points in this area Calculation of temperature difference in the region With the highest wall temperature in the region This is used for subsequent modeling and prediction.

[0045] Step 3: To address the issues of high dimensionality and redundancy in boiler input variables, and the different variables of interest for different objectives, this embodiment introduces a multi-objective random forest regression model (RF) for feature importance assessment and selection. The output objectives are multi-output sets: boiler efficiency, NOx, and several sets of regional temperature differences and maximum regional wall temperatures of water-cooled walls. To ensure data accuracy, rows containing missing values ​​are removed. A separate random forest regression model is trained for each output objective. The importance of out-of-bag (OOB) error perturbation is used as a variable contribution measure, and the top 10 input auxiliary variables corresponding to the importance indices for each output objective are selected as important features. The union of the importance results for each output objective is then used to obtain the final input feature set. This process ensures that the input features contribute significantly to multi-objective prediction while significantly reducing the model's input dimensionality and training complexity. The final input feature set is shown in Table 1 below.

[0046] Table 1 Step 4: Due to the significant differences in combustion mechanisms under different boiler operating conditions, this embodiment adopts a condition-based modeling strategy. Fuzzy C-means clustering is used to obtain membership degree representation for operating condition classification. Boiler load, total coal feed, and total air feed are selected as operating condition discriminant variables. The Euclidean distance between the standardized operating condition characteristics and the center of each cluster is calculated. The cluster with the smallest distance is taken as the current operating condition, and the corresponding operating condition combustion state prediction model is called to participate in the optimization calculation, thereby avoiding model mismatch caused by operating condition drift. After standardization, cluster analysis is performed to obtain K typical operating condition clusters (K=3).

[0047] Step 5: Within each operating condition cluster, this embodiment constructs a multi-output combustion state prediction model based on a CNN-BiLSTM network (Convolutional Neural Network-Bidirectional Long Short-Term Memory Network). To fully utilize the temporal correlation of the combustion process, a sliding time window is used to construct sequence samples: the window length is set to 30, and the input feature sequence of 30 consecutive time steps is used as the model input to predict the multidimensional output of the next time step. In terms of network structure, the model input layer is a sequence input layer (dimension is the number of input features p), followed by two one-dimensional convolutional modules to extract short-term fluctuations and local coupling response features of the combustion process, where the convolutional kernel sizes are 5 and 3, corresponding to 16 and 32 channels respectively. Convolutional features are input into a two-layer BiLSTM: the first BiLSTM adopts a sequence output mode, retaining the hidden state of each time step to capture the dependencies between time steps; the second BiLSTM adopts a final time step output mode, compressing and fusing the information of the entire time window. Subsequently, multidimensional output regression was achieved through a fully connected layer (FC=32, ReLU) and an output layer (FC=14) to jointly predict indicators such as boiler efficiency, NOx emission concentration, regional temperature difference of water-cooled walls, and regional maximum wall temperature. The prediction results are attached. Figures 2a to 2d As shown.

[0048] Step 6: After obtaining the combustion state prediction model for each operating condition, this embodiment constructs a multi-objective optimization decision system coupled with the combustion state prediction model. The control variables are selected as a set of operable combustion adjustment parameters, including: the opening degree of secondary dampers at each level (opening degree range 0-100), total air supply, and coal feed rate of each coal mill, etc. To consider the boiler control execution inertia and action feasibility, this embodiment introduces action holding constraints, that is, applying the same control action to multiple consecutive sampling points at the end of the window to simulate the execution process of the actual control system and avoid unexecutable solutions caused by instantaneous large changes. The multi-objective evaluation adopts a comprehensive objective function form, and the optimization directions include: reducing NOx emission concentration, improving boiler efficiency, reducing the maximum absolute temperature difference in the water-cooled wall area, and reducing the highest wall temperature in the water-cooled wall area, and setting action amplitude penalties to suppress excessive adjustment. To achieve the unification of different dimensions, this embodiment calculates a normalized scale based on the fluctuation statistics of the operating condition test section, and then performs a weighted summation after making the target changes dimensionless.

[0049] Step 7: During the optimization process, Particle Swarm Optimization (PSO) is used to search the high-dimensional control variable space. Each particle represents a set of candidate control action vectors, and its velocity and position are iteratively updated within the boundary constraints. Each iteration evaluates the particle fitness through the process of "action retention → constructing an adjusted window input → calling the combustion state prediction model for different operating conditions → calculating the comprehensive objective function", and updates the optimal solution accordingly.

[0050] Through the above implementation process, this application achieves the coupling of combustion prediction under different operating conditions and multi-objective optimization: the combustion state prediction model provides rapid evaluation capability for candidate actions, and the optimization algorithm searches for the optimal strategy under safety boundary constraints and action maintenance constraints, realizing the synergistic optimization of NOx emission concentration and boiler efficiency, while reducing the risk of temperature difference and maximum wall temperature in the water-cooled wall area, meeting the comprehensive requirements of boiler environmental protection, economy and safety under variable load operating conditions. Figures 3a to 3e As shown, after applying the optimization method of this application, under deep peak shaving operation conditions, the boiler efficiency was increased by an average of 0.019%, the NOx emission concentration at the SCR inlet was reduced by an average of 7.97 mg / m3, the average temperature difference of the water-cooled wall was reduced by 11.346℃, and the maximum wall temperature of the water-cooled wall was reduced by an average of 17.791℃, which proves the effectiveness of this application.

[0051] The above are exemplary embodiments of this application, and the scope of protection of this application is defined by the claims and their equivalents.

Claims

1. A combustion optimization method for a coal-fired boiler coupled with water-cooled wall temperature constraints, characterized in that, include: Step 1: Acquire historical operating data of the coal-fired boiler under different operating conditions; wherein, the historical operating data includes operating parameters and performance parameters, the operating parameters include boiler load, coal feed rate of each coal mill, total air supply, each air supply branch, opening degree of secondary air damper of each layer and oxygen content in flue gas, and the performance parameters include boiler efficiency, NOx emission concentration and water-cooled wall measuring point temperature; Step 2: Preprocess the historical running data to remove samples containing invalid data and obtain a continuous multivariate time series sample dataset; Step 3: Select boiler load, total coal feed and total air feed as operating condition discriminant variables from the historical operating data, perform cluster analysis on the feature vector formed by the operating condition discriminant variables, and divide the historical operating data into 3 typical operating conditions based on the results of the cluster analysis. Step 4: Using boiler efficiency, NOx emission concentration, regional temperature difference of water-cooled wall and regional maximum wall temperature as output targets, select corresponding input auxiliary variables from the multivariate time series sample dataset according to the output targets, perform feature filtering on the input auxiliary variables, and obtain the final input feature set; Step 5: Construct and train a combustion state prediction model for each typical operating condition; wherein, the combustion state prediction model is constructed using a combination of convolutional neural network and bidirectional long short-term memory network. Step 6: With the optimization objectives of reducing NOx emission concentration, improving boiler efficiency, reducing temperature difference in each water-cooled wall region, and reducing the maximum wall temperature of the water-cooled wall, construct the comprehensive objective function of the combustion state prediction model; Step 7: Within the safety boundary constraints, the control variable space is searched using a multi-objective intelligent optimization algorithm. The control variables are iteratively updated using the combustion state prediction model to minimize the comprehensive objective function, thereby obtaining the optimal control variable vector. The boiler combustion is then optimized and adjusted based on the optimal control variable vector. The safety boundary constraints are the minimum and maximum value boundaries corresponding to each parameter in the historical operating data. The control variables include the opening degree of each secondary damper, the total air supply, and the coal feed rate of each coal mill.

2. The method for optimizing combustion in a coal-fired boiler as described in claim 1, characterized in that, The clustering analysis employs the fuzzy C-means clustering method, and its objective function is expressed as: ； in, Indicates the first Feature vectors corresponding to historical operational data This represents the total number of samples in the historical operational data; Indicates the first Cluster centers corresponding to each operating condition category Indicates the number of operating condition categories; Representing the eigenvector For the Membership degree of each operating condition category; This represents the fuzziness index.

3. The method for optimizing combustion in a coal-fired boiler as described in claim 1, characterized in that, In step 4, feature filtering is performed on the input auxiliary variables to obtain the final input feature set, including: The input auxiliary variables are feature-selected using a random forest regression model. The random forest regression model calculates the importance index of the input auxiliary variables using out-of-bag error perturbation, as follows: ； in, Indicates the first Importance indicators of input auxiliary variables Indicates the number of decision trees. and These represent the out-of-bag data before and after the perturbation at the [number]th ... Prediction error on each decision tree; The importance indicators are sorted in descending order, and the top K importance indicators are selected. The input auxiliary variables corresponding to these top K importance indicators are the important features, and the union of these important features is the final set of input features.

4. The method for optimizing combustion in a coal-fired boiler as described in claim 1, characterized in that, The input sequence of the combustion state prediction model is: the continuous sequence in the final input feature set. If a time series is composed of characteristic variables at historical moments, then the time series is represented as: ； in, express The final input at any given time corresponds to the feature variable in the feature set. The output of the combustion state prediction model is expressed as follows: ； in and They represent The first water-cooled wall Temperature difference in each region and highest wall temperature in each region, express Boiler efficiency at any given time express NOx emission concentration at any given time.

5. The method for optimizing combustion in a coal-fired boiler as described in claim 1, characterized in that, The loss function of the combustion state prediction model is expressed as: ； in, Indicates the number of training samples. Represents the true value. This represents the predicted value output by the combustion state prediction model.

6. The method for optimizing combustion in a coal-fired boiler as described in claim 1, characterized in that, The comprehensive objective function is expressed as follows: ； ； in, Represents the overall objective function. This indicates that the action remains constrained. , , and These represent the changes in NOx emission concentration, boiler efficiency, temperature difference in each water-cooled wall region, and maximum water-cooled wall temperature relative to the current operating state under the action of candidate control actions. , , and These represent the statistical scales for NOx emission concentration, boiler efficiency, temperature difference in each water-cooled wall region, and maximum water-cooled wall temperature, respectively. , , , , All represent weighting coefficients; Indicates the first Candidate values ​​for each control variable; This indicates the value of the corresponding control variable under the current operating condition. This indicates the allowable range of variation for the control variable. This indicates the total number of control variables.

7. The method for optimizing combustion in a coal-fired boiler as described in claim 1, characterized in that, In step 2, the preprocessing method includes outlier removal and missing value processing.

8. A combustion optimization system for a coal-fired boiler coupled with water-cooled wall temperature constraints, the system being used to implement the combustion optimization method for a coal-fired boiler as described in any one of claims 1-7, characterized in that, The system includes: The data acquisition module acquires historical operating data of the coal-fired boiler under different operating conditions. The historical operating data includes operating parameters and performance parameters. The operating parameters include boiler load, coal feed rate of each coal mill, total air supply, air flow branches, secondary air damper opening of each layer, and flue gas oxygen content. The performance parameters include boiler efficiency, NOx emission concentration, and water-cooled wall measuring point temperature. The preprocessing module preprocesses the historical running data to remove samples containing invalid data, thereby obtaining a continuous multivariate time series sample dataset. The clustering module selects boiler load, total coal feed and total air feed as operating condition discriminant variables from the historical operating data, performs cluster analysis on the feature vector formed by the operating condition discriminant variables, and divides the historical operating data into 3 typical operating conditions based on the results of the cluster analysis. The feature filtering module takes boiler efficiency, NOx emission concentration, temperature difference of each water-cooled wall area and maximum water-cooled wall temperature as output targets, selects corresponding input auxiliary variables from the multivariate time series sample dataset according to the output targets, performs feature filtering on the input auxiliary variables, and obtains the final input feature set. The model building module constructs and trains a combustion state prediction model for each typical operating condition; wherein, the combustion state prediction model is constructed using a combination of convolutional neural network and bidirectional long short-term memory network. The objective construction module constructs the comprehensive objective function of the combustion state prediction model with the optimization objectives of reducing NOx emission concentration, improving boiler efficiency, reducing temperature difference in each water-cooled wall region, and reducing the maximum wall temperature of the water-cooled wall. The target optimization module searches the control variable space within the safety boundary constraints using a multi-objective intelligent optimization algorithm, iteratively updates the control variables using the combustion state prediction model to minimize the comprehensive objective function, thereby obtaining the optimal control variable vector, and then optimizes and adjusts boiler combustion based on the optimal control variable vector.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the coal-fired boiler combustion optimization method as described in any one of claims 1 to 7.

10. A computer storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the coal-fired boiler combustion optimization method as described in any one of claims 1 to 7.

Images