A coal-fired power plant boiler combustion optimization method based on a composite intelligent combustion model

Abstract

The present application relates to a kind of coal-fired power plant boiler combustion optimization method based on composite intelligent combustion model, comprising the following steps: step S1: the construction of composite intelligent combustion model;Step S2: complete boiler efficiency calculation and operation variable data arrangement, adopt artificial neural network ANN modeling, realize intelligent random optimization of combustion condition in combination with genetic algorithm GA;Step S3: result verification, whether artificial neural network ANN module can be verified in turn as the proxy model of genetic algorithm GA module optimization;Verify the boiler operating parameters optimized by genetic algorithm GA module;And verify the consistency of boiler operation data;Step S4: conclusion determination, by direct method and indirect method accounting, determine the numerical proportion of optimized boiler thermal efficiency, coal consumption value, and determine the optimal level of excess oxygen stable maintenance;The present application has the advantages that: with the maximization of boiler efficiency as the goal search global optimal operating parameter combination.For the same type unit provides replicable boiler thermal efficiency intelligent optimization scheme.

Description

Technical Field

[0001] This invention relates to the field of boiler combustion technology, and more specifically to a combustion optimization method for coal-fired power plant boilers based on a composite intelligent combustion model. Background Technology

[0002] The massive amount of boiler operation data accumulated by the distributed control system (DCS) of coal-fired power plants has provided new possibilities for data-driven combustion optimization. How to use artificial intelligence technology to mine the value of data and achieve precise control of boiler operating parameters has become the core research direction for the intelligent upgrading of coal-fired power plants.

[0003] The rapid development of artificial intelligence technology has provided a powerful tool for modeling and optimizing complex industrial systems. Against this backdrop, scholars both domestically and internationally have conducted extensive research on AI-based boiler combustion optimization. Lu Hongbo et al. proposed an improved multi-objective optimization method combining support vector machine modeling with genetic algorithm, particle swarm optimization, and differential evolution algorithm to address the problem of fly ash carbon content and NOx emissions in 300 MW coal-fired boilers. Verification showed that the model's fly ash carbon content verification error was 2.15% and NOx verification error was 1.24%, with the differential evolution algorithm showing the best optimization effect. Yu Tingfang et al. proposed an improved genetic algorithm combined with backpropagation neural network optimization method to address the low NOx combustion requirements of 300 MW power plant boilers, reducing NOx emissions by 35% compared to the average. Zheng Y et al. established a seven-step engineering method for coal-fired boiler combustion optimization based on artificial neural network and Nelder-Mead algorithm, achieving a 15% reduction in SCR inlet NOx emissions, a 12% reduction in ammonia consumption, and a 0.57% increase in unit thermal efficiency under full-load conditions. MVJJ Suresh et al. proposed a two-stage neural genetic optimization method coupling artificial neural network and genetic algorithm to achieve global optimization of operating parameters for coal-fired power plants, resulting in a power plant energy efficiency of 39.6% after optimization. The above research has laid a solid theoretical and technical foundation for combustion optimization of coal-fired boilers. However, traditional intelligent optimization methods have obvious limitations. First, they lack the ability to generalize operating conditions and are difficult to cover all operating conditions of the boiler. Second, the traditional ANN-GA coupling method has redundant input variables and low computational efficiency, making it difficult to achieve efficient and accurate combustion control. Summary of the Invention

[0004] In view of the above problems, the purpose of this invention is to provide a combustion optimization method for coal-fired power plant boilers based on a composite intelligent combustion model. This method initially selects boiler operating parameters based on the laws of energy and mass conservation. After eliminating redundant variables and selecting key influencing parameters using the PCA-PLS algorithm, a boiler efficiency prediction model based on a multilayer perceptron is constructed and trained and validated. This model is then used as the fitness function of a genetic algorithm to search for the globally optimal combination of operating parameters with the goal of maximizing boiler efficiency. This provides a replicable intelligent optimization scheme for boiler thermal efficiency for similar units, overcoming the shortcomings of the existing technologies.

[0005] This invention provides a combustion optimization method for coal-fired power plant boilers based on a composite intelligent combustion model, comprising the following steps: Step S1: Construction of the composite intelligent combustion model, including the construction of the principal component analysis (PCA) module, the partial least squares (PLS) module, the artificial neural network (ANN) module, and the genetic algorithm (GA) module; Step S2: Complete boiler efficiency calculation and operation variable data processing, use artificial neural network (ANN) modeling, and combine with genetic algorithm (GA) to achieve intelligent random optimization of combustion conditions; Step S3: Result verification. This involves verifying whether the Artificial Neural Network (ANN) module can serve as a proxy model for the Genetic Algorithm (GA) module in the optimization process; verifying the boiler operating parameters optimized by the GA module; and verifying the consistency of the boiler operating data. Step S4: Conclusion Confirmation. Through direct and indirect methods, the optimized boiler thermal efficiency ratio, coal consumption value, and the optimal level for maintaining excess oxygen stability are determined. This method can effectively screen key variables, accurately model, and achieve operating condition optimization, providing a replicable intelligent solution for similar units.

[0006] As a preferred embodiment of the present invention, in step S1, the principal component analysis (PCA) module and the partial least squares (PLS) module are used to remove redundant variables in the boiler operating parameters and screen the key parameters (principal components) that affect boiler efficiency, thereby reducing the complexity and computation time of the subsequent composite intelligent combustion model; then, the artificial neural network (ANN) module is used to construct a nonlinear mapping relationship between boiler efficiency and key operating parameters, thereby achieving high-precision prediction of boiler efficiency; finally, the genetic algorithm (GA) module uses the trained multilayer perceptron prediction model as the fitness function to search for the globally optimal combination of operating parameters within the parameter feasible region, thereby achieving global optimization of the boiler combustion process.

[0007] As a preferred embodiment of the present invention, step S1 further includes the following step: Step S11: Using the Principal Component Analysis (PCA) module, the data coordinate system is reconstructed through orthogonal transformation. The high-dimensional original variables with correlations are transformed into a few uncorrelated principal components, achieving dimensionality reduction while preserving key data information to the maximum extent. In the PCA transformation, the number of principal components is the same as the number of original variables, theoretically preserving all data information. Only core principal components are selected based on eigenvalues, eliminating redundant dimensions. The newly generated principal components are linear combinations of the original variables, possessing two core characteristics: first, the principal components are orthogonal and uncorrelated, completely eliminating multicollinearity interference between variables; second, the eigenvalues ​​corresponding to the principal components directly reflect their variance contribution rate. The larger the eigenvalue, the more important the data information contained in the corresponding principal component, which can be used to select key principal components. The principal components of the composite variables include: temperature, pressure, air volume, and coal quantity. Step S12: The Partial Least Squares (PLS) module is used to extract all independent variables from the principal components of the composite variables to predict the components of the response variable and establish the relationship between the response variable and the independent variables. The PLS module employs a multivariate linear model to connect the independent variable matrix and the dependent variable (response variable) matrix. Because PLS considers the structural characteristics of both the independent and dependent variable matrices, it can handle situations with a large number of independent variables, noise, multicollinearity, and incomplete data. Furthermore, the accuracy of the parameters extracted by PLS increases with the number of relevant variables and observations, making it suitable for modeling regression problems with one or more dependent variables using a series of predictor variables.

[0008] Step S13: Using the Artificial Neural Network (ANN) module, a Multilayer Perceptron (MLP) is used to receive the components of the response variable (key operating parameters), construct a nonlinear mapping relationship between boiler efficiency and the components of the response variable, and perform training. Step S14: Using the Genetic Algorithm (GA) module, firstly generate an initial population, then calculate the fitness value of each individual; the individual selection adopts an elite selection strategy, screening individuals based on their fitness values, with individuals having a higher fitness value having a greater probability of being selected; then, crossover operations are used to exchange some genes of parent individuals to generate offspring individuals, and then mutation operations are used to randomly change some genes of individuals to restore the genetic diversity of the population and avoid the algorithm getting trapped in local optima; finally, the fitness value of the new population, i.e., the fitness function, is calculated.

[0009] As a preferred embodiment of the present invention, step S2 further includes the following step: Step S21: Boiler efficiency calculation. Boiler thermal efficiency, as a core economic indicator characterizing the energy utilization level of the unit, is the core objective function of combustion optimization. The calculation of the thermal efficiency of coal-fired power plant boilers includes the input-output method (direct method) and / or the heat loss method (indirect method). Step S211: The input-output method calculates the boiler thermal efficiency using the ratio of energy contained in water and steam to that contained in fuel; the calculation formula is as follows: ; In the formula, Represents boiler thermal efficiency, % Represents steam mass flow rate, kg / h; Represents fuel mass flow rate, kg / h; Represents vapor enthalpy, kJ / kg; Represents the enthalpy of water supply, kJ / kg; This represents the higher heating value of the fuel, in kJ / kg.

[0010] Step S212: The heat loss method determines the boiler thermal efficiency by calculating the difference between the input energy and various heat losses. The data used for calculating the boiler thermal efficiency using the heat loss method includes boiler operating data and fuel analysis data. The boiler operating data includes fuel mass flow rate, steam mass flow rate, steam outlet pressure and temperature, feedwater temperature, CO2 and CO content in flue gas, exhaust gas temperature, ambient temperature, ambient air humidity, boiler surface temperature, air mass flow rate, total boiler surface area, and combustible content of bottom ash and fly ash. The fuel analysis data includes the ash, carbon, hydrogen, nitrogen, oxygen, and sulfur content of the fuel, as well as the higher heating value.

[0011] As a preferred embodiment of the present invention, step S2 further includes the following step: Step S22: Obtain boiler parameter and variable data. Step S221: Obtain the operating data of the coal-fired power plant boiler through the DCS system (distributed control system), wherein the operating data of the coal-fired power plant boiler is used as training samples for the artificial neural network (ANN) module. Step S222: Select boiler operating parameters as input parameters and boiler thermal efficiency as output parameters; Step S223: Use the partial least squares (PLS) module to reduce the dimensionality of the independent variables in the acquired coal-fired power plant boiler operation data to response variables.

[0012] As a preferred embodiment of the present invention, step S2 further includes the following step: Step S23: Model the data of the Artificial Neural Network (ANN) module and determine the fitness function; Step S231: The multilayer perceptron (MLP) used in the artificial neural network (ANN) module is a multilayer neural network composed of an input layer, a hidden layer, and an output layer; Step S232: Input the input parameters to be optimized into a multilayer perceptron (MLP) for training, and its output is used as the boiler efficiency. The input parameters include CF, PAF, SAF, FWF, FWE, SF, SE, LOAD, EA, MOT, and BTB.

[0013] As a preferred embodiment of the present invention, step S2 further includes the following step: Step S24: Based on the stochastic optimization of the Genetic Algorithm (GA) module, determine the objective function and fitness function, and solve for the optimal values ​​of the boiler operating parameters; wherein, the objective function of the Genetic Algorithm (GA) module is to maximize the boiler efficiency while controlling the excess oxygen content to ensure that the excess oxygen content is still lower than the set value when the generator reaches its maximum output.

[0014] As a preferred embodiment of the present invention, step S3 further includes the following step: Step S31: The number of hidden nodes in an ANN is a key parameter affecting the model's fitting performance and generalization ability. Too few nodes can lead to underfitting and insufficient prediction accuracy, while too many nodes can cause overfitting and increase computational overhead. Therefore, it is necessary to optimize the number of nodes using quantitative indicators. The optimal number of nodes is determined by observing the root mean square error (RMSE) under different numbers of hidden nodes in the ANN module; and the determination coefficients of the training and validation sets are used to determine whether overfitting has occurred. Step S32: Using the trained and validated Artificial Neural Network (ANN) module as a surrogate model, the Genetic Algorithm (GA) module is used to optimize the boiler operating parameters, with the goal of maximizing boiler thermal efficiency, to solve for the optimal operating conditions of each input variable; based on the optimization results, the boiler performance before and after optimization is compared and analyzed. Step S33: Verify the consistency of boiler operating data based on the deviation between the input-output method and the heat loss method.

[0015] The advantages and positive effects of this invention are as follows: Addressing the problems of variable redundancy and computational inefficiency in traditional intelligent optimization methods for pulverized coal boilers in coal-fired power plants, this application proposes a composite intelligent method based on principal component analysis-partial least squares (PCA-PLS) variable selection and artificial neural network-genetic algorithm (ANN-GA). Based on the laws of energy and mass conservation, boiler operating parameters are initially selected. After redundant variables are eliminated and key influencing parameters are selected using the PCA-PLS algorithm, a boiler efficiency prediction model based on a multilayer perceptron is constructed and trained and validated. This model is then used as the fitness function of the genetic algorithm to search for the globally optimal combination of operating parameters with the goal of maximizing boiler efficiency. The research results can provide replicable intelligent optimization schemes for similar units and have significant engineering application value for promoting the efficient, green, and low-carbon operation of coal-fired power plants in my country. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the multilayer perceptron (MLP) structure in this invention.

[0017] Figure 2 This is a schematic diagram of the boiler operating parameter settings in this invention.

[0018] Figure 3 This is a schematic diagram of the PCA analysis results in this invention.

[0019] Figure 4 This is a schematic diagram of the PLS analysis results in this invention.

[0020] Figure 5 This is a schematic diagram of the ANN architecture in this invention.

[0021] Figure 6 This is a schematic diagram of the root mean square error results in this invention.

[0022] Figure 7 This is a schematic diagram illustrating the performance metrics of ANN model training and validation in this invention.

[0023] Figure 8 This is a schematic diagram of the iterative convergence curve of the GA algorithm in this invention.

[0024] Figure 9 This is a schematic diagram comparing the boiler thermal efficiency before and after optimization in this invention.

[0025] Figure 10 This is a schematic diagram comparing the boiler operating parameter values ​​before and after optimization in this invention. Detailed Implementation

[0026] In the following description, numerous specific details are set forth for illustrative purposes and to provide a thorough understanding of one or more embodiments. However, it will be apparent that these embodiments may also be implemented without these specific details. In other instances, well-known structures and devices are shown in block diagram form for ease of description of one or more embodiments.

[0027] Example 1

[0028] like Figure 1 As shown in the figure, an embodiment of the present invention provides a combustion optimization method for coal-fired power plant boilers based on a composite intelligent combustion model, comprising the following steps: Step S1: Construction of the composite intelligent combustion model, including the construction of Principal Component Analysis (PCA) module, Partial Least Squares (PLS) module, Artificial Neural Network (ANN) module, and Genetic Algorithm (GA) module. The PCA and PLS modules are used to eliminate redundant variables in the boiler operating parameters and screen key parameters (principal components) that affect boiler efficiency, reducing the complexity and computation time of the subsequent composite intelligent combustion model. Then, the ANN module is used to construct a nonlinear mapping relationship between boiler efficiency and key operating parameters, achieving high-precision prediction of boiler efficiency. Finally, the GA module uses a trained multilayer perceptron prediction model as the fitness function to search for the globally optimal combination of operating parameters within the parameter feasible region, achieving global optimization of the boiler combustion process.

[0029] Step S11: Using the Principal Component Analysis (PCA) module, the data coordinate system is reconstructed through orthogonal transformation. The high-dimensional original variables with correlations are transformed into a few uncorrelated principal components, achieving dimensionality reduction while preserving key data information to the maximum extent. In the PCA transformation, the number of principal components is the same as the number of original variables, theoretically preserving all data information. Only core principal components are selected based on eigenvalues, eliminating redundant dimensions. The newly generated principal components are linear combinations of the original variables, possessing two core characteristics: first, the principal components are orthogonal and uncorrelated, completely eliminating multicollinearity interference between variables; second, the eigenvalues ​​corresponding to the principal components directly reflect their variance contribution rate. The larger the eigenvalue, the more important the data information contained in the corresponding principal component, which can be used to select key principal components. The principal components of the composite variables include: temperature, pressure, air volume, and coal quantity. Step S12: The Partial Least Squares (PLS) module is used to extract all independent variables from the principal components of the composite variables to predict the components of the response variable and establish the relationship between the response variable and the independent variables. The PLS module employs a multivariate linear model to connect the independent variable matrix and the dependent variable (response variable) matrix. Because PLS considers the structural characteristics of both the independent and dependent variable matrices, it can handle situations with a large number of independent variables, noise, multicollinearity, and incomplete data. Furthermore, the accuracy of the parameters extracted by PLS increases with the number of relevant variables and observations, making it suitable for modeling regression problems with one or more dependent variables using a series of predictor variables.

[0030] Step S13: Using the Artificial Neural Network (ANN) module, a Multilayer Perceptron (MLP) is used to receive the components of the response variable (key operating parameters), construct a nonlinear mapping relationship between boiler efficiency and the components of the response variable, and perform training. Step S14: Using the Genetic Algorithm (GA) module, firstly generate an initial population, then calculate the fitness value of each individual; the individual selection adopts an elite selection strategy, screening individuals based on their fitness values, with individuals having a higher fitness value having a greater probability of being selected; then, crossover operations are used to exchange some genes of parent individuals to generate offspring individuals, and then mutation operations are used to randomly change some genes of individuals to restore the genetic diversity of the population and avoid the algorithm getting trapped in local optima; finally, the fitness value of the new population, i.e., the fitness function, is calculated.

[0031] Step S2: Complete boiler efficiency calculation and operation variable data processing, use artificial neural network (ANN) modeling, and combine with genetic algorithm (GA) to achieve intelligent random optimization of combustion conditions; Step S21: Boiler efficiency calculation. Boiler thermal efficiency, as a core economic indicator characterizing the energy utilization level of the unit, is the core objective function of combustion optimization. The calculation of the thermal efficiency of coal-fired power plant boilers includes the input-output method (direct method) and / or the heat loss method (indirect method). Step S211: The input-output method calculates the boiler thermal efficiency using the ratio of energy contained in water and steam to that contained in fuel; the calculation formula is as follows: ; In the formula, Represents boiler thermal efficiency, % Represents steam mass flow rate, kg / h; Represents fuel mass flow rate, kg / h; Represents vapor enthalpy, kJ / kg; Represents the enthalpy of water supply, kJ / kg; This represents the higher heating value of the fuel, in kJ / kg.

[0032] Step S212: The heat loss method determines the boiler thermal efficiency by calculating the difference between the input energy and various heat losses. The data used for calculating the boiler thermal efficiency using the heat loss method includes boiler operating data and fuel analysis data. The boiler operating data includes fuel mass flow rate, steam mass flow rate, steam outlet pressure and temperature, feedwater temperature, CO2 and CO content in flue gas, exhaust gas temperature, ambient temperature, ambient air humidity, boiler surface temperature, air mass flow rate, total boiler surface area, and combustible content of bottom ash and fly ash. The fuel analysis data includes the ash, carbon, hydrogen, nitrogen, oxygen, and sulfur content of the fuel, as well as the higher heating value.

[0033] Step S22: Obtain boiler parameter and variable data. Step S221: Obtain the operating data of the coal-fired power plant boiler through the DCS system (distributed control system), wherein the operating data of the coal-fired power plant boiler is used as training samples for the artificial neural network (ANN) module. Step S222: Select boiler operating parameters as input parameters and boiler thermal efficiency as output parameters; Step S223: Use the partial least squares (PLS) module to reduce the dimensionality of the independent variables in the acquired coal-fired power plant boiler operation data to response variables.

[0034] Step S23: Model the data of the Artificial Neural Network (ANN) module and determine the fitness function; Step S231: The multilayer perceptron (MLP) used in the artificial neural network (ANN) module is a multilayer neural network composed of an input layer, a hidden layer, and an output layer; Step S232: Input the input parameters to be optimized into a multilayer perceptron (MLP) for training, and its output is used as the boiler efficiency. The input parameters include CF, PAF, SAF, FWF, FWE, SF, SE, LOAD, EA, MOT, and BTB.

[0035] Step S24: Based on the stochastic optimization of the Genetic Algorithm (GA) module, determine the objective function and fitness function, and solve for the optimal values ​​of the boiler operating parameters; wherein, the objective function of the Genetic Algorithm (GA) module is to maximize the boiler efficiency while controlling the excess oxygen content to ensure that the excess oxygen content is still lower than the set value when the generator reaches its maximum output.

[0036] Step S3: Result verification. This involves verifying whether the Artificial Neural Network (ANN) module can serve as a proxy model for the Genetic Algorithm (GA) module in the optimization process; verifying the boiler operating parameters optimized by the GA module; and verifying the consistency of the boiler operating data. Step S31: The number of hidden nodes in an ANN is a key parameter affecting the model's fitting performance and generalization ability. Too few nodes can lead to underfitting and insufficient prediction accuracy, while too many nodes can cause overfitting and increase computational overhead. Therefore, it is necessary to optimize the number of nodes using quantitative indicators. The optimal number of nodes is determined by observing the root mean square error (RMSE) under different numbers of hidden nodes in the ANN module; and the determination coefficients of the training and validation sets are used to determine whether overfitting has occurred. Step S32: Using the trained and validated Artificial Neural Network (ANN) module as a surrogate model, the Genetic Algorithm (GA) module is used to optimize the boiler operating parameters, with the goal of maximizing boiler thermal efficiency, to solve for the optimal operating conditions of each input variable; based on the optimization results, the boiler performance before and after optimization is compared and analyzed. Step S33: Verify the consistency of boiler operating data based on the deviation between the input-output method and the heat loss method.

[0037] Step S4: Conclusion Confirmation. Through direct and indirect methods, the optimized boiler thermal efficiency ratio, coal consumption value, and the optimal level for maintaining excess oxygen stability are determined. This method can effectively screen key variables, accurately model, and achieve operating condition optimization, providing a replicable intelligent solution for similar units.

[0038] Furthermore, the Principal Component Analysis (PCA) module in this embodiment is the most classic and widely used linear dimensionality reduction algorithm in the field of multivariate statistical analysis. Its core idea is to reconstruct the data coordinate system through orthogonal transformation, transforming the correlated high-dimensional original variables into a few uncorrelated composite variables (i.e., principal components), thereby achieving dimensionality reduction while preserving key data information to the maximum extent. In the PCA transformation, the number of principal components is the same as the number of original variables, theoretically preserving all data information. Only core principal components are selected through eigenvalues, eliminating redundant dimensions. The newly generated principal components are linear combinations of the original variables, possessing two core characteristics: first, the principal components are mutually orthogonal and uncorrelated, completely eliminating multicollinearity interference among variables; second, the eigenvalues ​​corresponding to the principal components directly reflect their variance contribution rate. The larger the eigenvalue, the more important the data information contained in the corresponding principal component, which can be used to select key principal components.

[0039] Furthermore, the Partial Least Squares (PLS) module in this embodiment is a second-generation multivariate statistical modeling method that integrates the core ideas of PCA, canonical correlation analysis, and multiple linear regression. It is specifically designed to address the challenges of high dimensionality, multicollinearity, small sample sizes, and noise interference that traditional regression methods struggle to handle. The PLS module reduces a large number of independent variables to a few new, uncorrelated variables that can well explain data variation. It then constructs new independent variables called factors, latent variables, or components (each component is a linear combination of the original independent variables) to establish the relationship between the response variable and the independent variables. The core objective of the PLS module is to extract components that best capture the information of the independent variables and are used to predict the response variable. This method connects two data matrices (the independent variable matrix and the dependent variable matrix) through a multiple linear model, thus considering the structural characteristics of both matrices. The advantage of the PLS module lies in its ability to handle situations with a large number of independent variables, noise, multicollinearity, and incomplete data. Furthermore, the parameter accuracy of the Partial Least Squares (PLS) module increases with the number of relevant variables and observations, making it suitable for modeling regression problems involving one or more dependent variables through a series of predictor variables.

[0040] Furthermore, the Artificial Neural Network (ANN) module in this embodiment is a continuous nonlinear dynamic system composed of a large number of simple processing units connected by weighted links. It can autonomously learn the complex implicit mapping relationship between input and output through data-driven methods, without needing to pre-establish an analytical mathematical model of the system. This characteristic gives it an unparalleled advantage over traditional mechanistic modeling methods in handling industrial process modeling problems with strong nonlinearity, strong coupling, and multiple variables. The ANN module consists of a large number of processing nodes, which are similar to neurons in the brain. The ANN module mainly consists of two parts: processing units (neurons) and the connections between units. The topology of the ANN is a multilayer perceptron (MLP). Figure 1 An ANN with an MLP structure is shown, and its mathematical expression is as follows.

[0041] ; In the formula, Representing the One output variable; Represents the activation function of the output layer; This represents the number of neurons in the hidden layer; Represents the connection weights from the hidden layer to the output layer; Representing the One hidden layer activation function; Represents the number of input variables; Represents the connection weights from the input layer to the hidden layer; Representing the One input variable; Represents the hidden layer bias; This represents the output layer bias.

[0042] To determine the values ​​of the weights W, a sufficient amount of interrelated input and output data is required. The process of determining the weights is called training or learning. The goal of ANN training is to minimize the error between the network output and the actual process output.

[0043] Furthermore, the core elements of the Genetic Algorithm (GA) module in this embodiment include: population, fitness function, selection, crossover, and mutation operations. The fitness function quantifies the quality of individuals and is the sole criterion driving evolution. The GA algorithm's operation typically includes: first, randomly initializing the population; then entering an iterative loop, calculating the fitness of each individual, performing the selection to retain dominant individuals, and generating a new generation of population through crossover and mutation; finally, determining whether the termination condition is met (such as reaching the maximum number of iterations or fitness convergence). Thanks to its population search mechanism and probabilistic operations, GA can effectively handle high-dimensional, multi-peak, nonlinear, and discontinuous multivariate optimization problems. Since it relies only on fitness values ​​rather than gradient information, it does not require the differentiability of the objective function, and its inherent parallel search characteristics enable it to effectively escape local optima traps.

[0044] Example 2 In this embodiment, 9132 sets of historical data under stable operating conditions were selected to establish a composite intelligent combustion model. The data was randomly divided, with 80% (7306 sets) of samples used for model training and the remaining 20% ​​(1826 sets) of samples used for model validation.

[0045] Based on the laws of conservation of energy and mass, 14 boiler operating parameters were selected as model inputs, and boiler thermal efficiency was used as the model output. (See reference...) Figure 2 As shown, the operating parameters are first analyzed and screened using the PCA-PLS method (Principal Component Analysis PCA module and Partial Least Squares PLS module), and then used for artificial neural network (ANN) module modeling and genetic algorithm (GA) module optimization.

[0046] The correlation coefficient was used to analyze the degree of association between the variables. Principal component analysis (PCA) and partial least squares (PLS) were used to screen the response variables that had the most significant impact on the combustion process and boiler efficiency for subsequent ANN modeling.

[0047] See Figure 3 PCA analysis results showed that burner angles A, B, C, and D formed a cluster. These variables were highly correlated and close to each other, indicating that the variables in this cluster were highly representative and could replace other variables in subsequent processing. From this cluster, a variable with clear physical meaning and easy measurement was selected, namely burner angle B (BTB), and the remaining burner angle variables were no longer included in subsequent calculations.

[0048] See Figure 4The PLS analysis results show that the projected importance values ​​of all variables are greater than 0.5, indicating that these variables are all significant and should be included in the ANN modeling. After screening using the PCA-PLS method, the variables used for ANN modeling are: coal flow rate, primary air flow rate, secondary air flow rate, feedwater flow rate, feedwater enthalpy, steam flow rate, steam enthalpy, unit load, excess oxygen, coal mill outlet temperature, burner tilt angle B, and boiler efficiency.

[0049] See Figure 5 This study uses an ANN-based data model, employing an ANN with a multilayer perceptron (MLP) structure as the model for the system to be optimized, and trains it using the Levenberg-Marquardt algorithm. Two key considerations during the modeling process are the selection of input and output variables and the determination of the number of hidden layer nodes. The selection of input and output variables must be based on the correlation between them; inappropriate selection will lead to a decrease in model accuracy. Similarly, determining the number of hidden layer nodes is crucial; too few or too many nodes will negatively impact the ANN's prediction accuracy.

[0050] Before optimization, modeling is first performed using an ANN: the operating parameters to be optimized (CF, PAF, SAF, FWF, FWE, SF, SE, LOAD, EA, MOT, BTB) are taken as input, and boiler efficiency is taken as output. The ANN architecture used is as follows: Figure 5 As shown, this is a multi-layer neural network containing an input layer, a hidden layer, and an output layer, trained using the Levenberg-Marquardt algorithm.

[0051] Stochastic optimization based on GA (General Aspect-Oriented Optimization) uses the GA algorithm to find the optimal values ​​of boiler operating parameters. The design of the GA algorithm requires determining the objective function and fitness function; even small errors in either function can lead to poor optimization results. The objective function of the GA algorithm is to maximize boiler efficiency while controlling excess oxygen content, ensuring that the excess oxygen content remains below the set value even when the generator reaches maximum output.

[0052] The optimization steps of the ANN-GA joint algorithm are as follows: First, an initial population is generated, and then the fitness value of each individual is calculated. Individual selection adopts an elite selection strategy, which selects individuals based on their fitness values. Individuals with higher fitness have a greater probability of being selected. Then, crossover operations are used to exchange some genes of parent individuals to generate offspring individuals. Mutation operations are then used to randomly change some genes of individuals to restore the genetic diversity of the population and avoid the algorithm from getting trapped in local optima. Finally, the fitness value of the new population is calculated.

[0053] Results verification: The optimal number of nodes was determined by observing the root mean square error (RMSE) under different numbers of hidden nodes. (See [reference needed]) Figure 6The curves showing the RMSE value changing with the number of hidden nodes and the root mean square error (RMSE) results were obtained. The graphs show that the RMSE value is minimized when the number of hidden nodes is 12, thus determining the optimal hidden layer configuration for the model. Based on this, the network topology was determined to be: 11 input nodes, 12 hidden nodes, and 1 output node. The statistical results of the model's prediction performance on the training and validation sets are as follows: Figure 7 As shown in the evaluation metrics, the determination coefficients of both the training and validation sets are higher than 0.985, the RMSE values ​​of the two sets of samples are very small, the model has no significant overfitting phenomenon, and the generalization performance is excellent. It can be used as a surrogate model for subsequent GA optimization.

[0054] See Figure 7 and Figure 8 Using a neural network that has undergone training and validation as a surrogate model, the GA algorithm is employed to optimize boiler operating parameters, with the goal of maximizing boiler thermal efficiency. The optimal operating conditions for each input variable are then determined. The optimal fitness value of the GA algorithm corresponds to a boiler thermal efficiency of 89.58%, which is the optimal boiler efficiency calculated using the indirect method.

[0055] See Figure 9 Based on the optimization results, a comparative analysis of the boiler performance before and after optimization was conducted. The deviations between the calculation results of the direct method and the indirect method were 1.57% and 1.63%, respectively, verifying the consistency of the boiler operating data. The efficiency of the direct method is slightly lower than that of the indirect method, mainly because the direct method is affected by steam flow measurement errors and heat loss in the pipeline, while the indirect method is based on heat loss calculation item by item, resulting in relatively higher accuracy.

[0056] See Figure 10 The input parameter values ​​can effectively improve boiler efficiency and reduce production costs by reducing coal consumption. Before optimization, the boiler coal flow rate was 125,927.20 kg / h at a load of 224.87 MW; after optimization, the boiler coal flow rate was only 122,967.91 kg / h at a load of 225.12 MW, which can reduce coal consumption by 2,959.29 kg per hour. To achieve this benefit, the excess oxygen needs to be maintained at the optimal value of 1.25%.

[0057] The conclusion confirms that the constructed composite intelligent model combining PCA-PLS and ANN-GA can be applied to the combustion optimization of coal-fired power plant boilers. Calculations using both direct and indirect methods show that the optimized boiler thermal efficiency is 88.00% and 89.58%, respectively, with coal consumption reduced by 2959.29 kg / h and excess oxygen maintained at an optimal level. This model can effectively screen key variables, accurately model, and achieve operating condition optimization, providing a replicable intelligent solution for similar units.

[0058] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for optimizing the combustion of a coal-fired power plant boiler based on a composite intelligent combustion model, characterized in that, Includes the following steps: Step S1: Construction of the composite intelligent combustion model, including the construction of the principal component analysis (PCA) module, the partial least squares (PLS) module, the artificial neural network (ANN) module, and the genetic algorithm (GA) module; Step S2: Complete boiler efficiency calculation and operation variable data processing, use artificial neural network (ANN) modeling, and combine with genetic algorithm (GA) to achieve intelligent random optimization of combustion conditions; Step S3: Result verification. This involves verifying whether the Artificial Neural Network (ANN) module can serve as a proxy model for the Genetic Algorithm (GA) module in the optimization process; verifying the boiler operating parameters optimized by the GA module; and verifying the consistency of the boiler operating data. Step S4: Conclusion confirmed. Through direct and indirect methods, the optimized boiler thermal efficiency ratio, coal consumption value, and the optimal level for maintaining excess oxygen stability are determined.

2. The combustion optimization method for coal-fired power plant boilers based on a composite intelligent combustion model according to claim 1, characterized in that, In step S1, the principal component analysis (PCA) module and the partial least squares (PLS) module are used to remove redundant variables in the boiler operating parameters and screen key parameters that affect boiler efficiency, thereby reducing the complexity and computation time of the subsequent composite intelligent combustion model. Then, the artificial neural network (ANN) module is used to construct a nonlinear mapping relationship between boiler efficiency and key operating parameters, achieving high-precision prediction of boiler efficiency. Finally, the genetic algorithm (GA) module uses the trained multilayer perceptron prediction model as the fitness function to search for the globally optimal combination of operating parameters within the parameter feasible region, thereby achieving global optimization of the boiler combustion process.

3. The combustion optimization method for coal-fired power plant boilers based on a composite intelligent combustion model according to claim 2, characterized in that, Step S1 also includes the following steps: Step S11: Using the Principal Component Analysis (PCA) module, the data coordinate system is reconstructed through orthogonal transformation. The high-dimensional original variables with correlations are transformed into a few uncorrelated principal components of comprehensive variables. This achieves dimensionality reduction while preserving key information of the data to the maximum extent. The principal components of comprehensive variables include: temperature, pressure, air volume, and coal volume. Step S12: Using the Partial Least Squares (PLS) module, extract all independent variables from the principal components of the comprehensive variable to predict the components of the response variable and establish the relationship between the response variable and the independent variables. Step S13: Use the Artificial Neural Network (ANN) module to receive the components of the response variable through a Multilayer Perceptron (MLP) to construct a nonlinear mapping relationship between boiler efficiency and the components of the response variable, and then train the module. Step S14: Using the Genetic Algorithm (GA) module, firstly generate an initial population, then calculate the fitness value of each individual; the individual selection adopts an elite selection strategy, screening individuals based on their fitness values, with individuals having a higher fitness value having a greater probability of being selected; then, crossover operations are used to exchange some genes of parent individuals to generate offspring individuals, and then mutation operations are used to randomly change some genes of individuals to restore the genetic diversity of the population and avoid the algorithm getting trapped in local optima; finally, the fitness value of the new population, i.e., the fitness function, is calculated.

4. The combustion optimization method for coal-fired power plant boilers based on a composite intelligent combustion model according to claim 1, characterized in that, Step S2 also includes the following steps: Step S21: Boiler efficiency calculation. Boiler thermal efficiency, as a core economic indicator characterizing the energy utilization level of the unit, is the core objective function of combustion optimization. The calculation of boiler thermal efficiency in coal-fired power plants includes the input-output method and / or the heat loss method. Step S211: The input-output method calculates the boiler thermal efficiency using the ratio between the energy contained in water and steam and that contained in fuel; Step S212: The heat loss method determines the boiler thermal efficiency by calculating the difference between the input energy and various heat losses. The data used for calculating the boiler thermal efficiency using the heat loss method includes boiler operating data and fuel analysis data. The boiler operating data includes fuel mass flow rate, steam mass flow rate, steam outlet pressure and temperature, feedwater temperature, CO2 and CO content in flue gas, exhaust gas temperature, ambient temperature, ambient air humidity, boiler surface temperature, air mass flow rate, total boiler surface area, and combustible content of bottom ash and fly ash. The fuel analysis data includes the ash, carbon, hydrogen, nitrogen, oxygen, and sulfur content of the fuel, as well as the higher heating value.

5. The combustion optimization method for coal-fired power plant boilers based on a composite intelligent combustion model according to claim 4, characterized in that, Step S2 also includes the following steps: Step S22: Obtain boiler parameter and variable data. Step S221: Obtain the boiler operation data of the coal-fired power plant through the DCS system, wherein the boiler operation data of the coal-fired power plant is used as the training sample of the artificial neural network (ANN) module; Step S222: Select boiler operating parameters as input parameters and boiler thermal efficiency as output parameters; Step S223: Use the partial least squares (PLS) module to reduce the dimensionality of the independent variables in the acquired coal-fired power plant boiler operation data to response variables.

6. The combustion optimization method for coal-fired power plant boilers based on a composite intelligent combustion model according to claim 5, characterized in that, Step S2 also includes the following steps: Step S23: Model the data of the Artificial Neural Network (ANN) module and determine the fitness function; Step S231: The multilayer perceptron (MLP) used in the artificial neural network (ANN) module is a multilayer neural network composed of an input layer, a hidden layer, and an output layer; Step S232: Input the input parameters to be optimized into a multilayer perceptron (MLP) for training, and its output is used as the boiler efficiency.

7. The combustion optimization method for coal-fired power plant boilers based on a composite intelligent combustion model according to claim 6, characterized in that, Step S2 also includes the following steps: Step S24: Based on the stochastic optimization of the Genetic Algorithm (GA) module, determine the objective function and fitness function, and solve for the optimal values ​​of the boiler operating parameters; wherein, the objective function of the Genetic Algorithm (GA) module is to maximize the boiler efficiency while controlling the excess oxygen content to ensure that the excess oxygen content is still lower than the set value when the generator reaches its maximum output.

8. The combustion optimization method for coal-fired power plant boilers based on a composite intelligent combustion model according to claim 1, characterized in that, Step S3 also includes the following steps: Step S31: Determine the optimal number of nodes by observing the root mean square error under different numbers of hidden nodes in the artificial neural network (ANN) module; and determine whether overfitting occurs based on the determination coefficients of the training and validation sets. Step S32: Using the trained and validated Artificial Neural Network (ANN) module as a surrogate model, the Genetic Algorithm (GA) module is used to optimize the boiler operating parameters, with the goal of maximizing boiler thermal efficiency, to solve for the optimal operating conditions of each input variable; based on the optimization results, the boiler performance before and after optimization is compared and analyzed. Step S33: Verify the consistency of boiler operating data based on the deviation between the input-output method and the heat loss method.

Images