A multi-objective optimization method for coal-fired boilers based on evolutionary algorithms

Abstract

The present invention discloses a multi-objective optimization method for coal-fired boilers based on an evolutionary algorithm. The present invention comprehensively considers multiple operating variables for optimizing multiple target variables. By establishing a gradient descent decision tree model for each target parameter to simulate the boiler combustion process, and on the basis of establishing the boiler combustion model, combining the NSGA-II algorithm to iteratively optimize the key parameters in the key parameter set, the optimization of multi-objective parameters related to improving the boiler combustion efficiency and reducing pollutant emissions is realized, so as to achieve the goal of improving the combustion efficiency and reducing pollutant emissions. The present invention has a high optimization rate for coal-fired boilers and can realize online adjustment of operating variables according to boiler operation data to achieve the goal of improving the combustion efficiency and reducing pollutant emissions.

Description

Technical Field

[0001] The present invention relates to the technical field of coal combustion optimization, and more specifically to a multi-objective optimization method for coal-fired boilers based on an evolutionary algorithm. Background Art

[0002] In the short term, thermal power generation remains the most important part of China's power resource structure. Under the dual pressures of reduced profit margins and increased environmental protection requirements, thermal power plants need to enhance their competitiveness by improving boiler combustion efficiency and reducing pollutant emissions. For thermal power units, there are many operating variables involved in their combustion process. In actual production, it is usually manually adjusted based on personal experience, and it is difficult to find the optimal operating conditions. At the same time, with the continuous development of artificial intelligence algorithms, the use of data-driven methods to solve boiler combustion optimization problems has received extensive attention. Therefore, establishing a multi-input multi-output model of the boiler combustion process, using a multi-objective evolutionary algorithm to optimize and adjust based on this model, obtaining the optimal control parameters under the current working conditions, and improving combustion efficiency and reducing pollutant emissions are issues that are currently relatively concerned. Summary of the Invention

[0003] The object of the present invention is to provide a multi-objective optimization method for coal-fired boilers based on an evolutionary algorithm. This method comprehensively considers multiple operating variables to optimize multiple target variables, and realizes the optimization of multi-objective parameters related to improving boiler combustion efficiency and reducing pollutant emissions on the basis of establishing a boiler combustion model.

[0004] To achieve the above object of the invention, the technical solution proposed by the present invention is as follows:

[0005] A multi-objective optimization method for coal-fired boilers based on an evolutionary algorithm, which includes the following steps:

[0006] S1. Collect historical data of the operating parameter set and the target parameter set during the operation of the coal-fired boiler. The operating parameter set includes two categories: non-adjustable operating parameters and adjustable operating parameters; divide the historical data into a training set and a test set, and perform dimensionality reduction on the operating parameter sets included in the samples in the training set and the test set, and retain the key parameter set that can affect the target parameters;

[0007] S2. For each target parameter y in the target parameter set, train a boiler combustion process model using the training set respectively, and verify it with the test set; the boiler combustion process model is a gradient descent decision tree model with the key parameter set as the input and the target parameter y as the output;

[0008] S3, set the objective function and constraint conditions according to the target parameter set of the coal-fired boiler, obtain the key parameter set at the current moment in real time during the operation of the coal-fired boiler to initialize the population, and iterate and optimize the key parameters in the key parameter set through the NSGA-Ⅱ algorithm. In each iterative process, the objective function value corresponding to the population individual is calculated through the boiler combustion process model trained in S2; after the optimization is completed, the Pareto optimal solution set is obtained;

[0009] S4. Select again from the Pareto optimal solutions obtained in S3 to obtain an optimal solution with the highest degree of optimization, and adjust the operating parameters of the coal-fired boiler according to this optimal solution.

[0010] Preferably, in step S1, the following steps are specifically included:

[0011] S101. According to a preset operating parameter set and a target parameter set, historical data of each operating parameter and target parameter in the operation process of the coal-fired boiler are collected; the target parameter set includes a boiler combustion efficiency index and a pollutant emission index, and the operating parameter set includes all non-adjustable operating parameters and adjustable operating parameters related to any target parameter in the coal-fired boiler;

[0012] S102, preprocessing the historical data, removing outliers, and then dividing the historical data into a training set and a test set, wherein each sample in the training set and the test set includes an operating parameter set and a target parameter set at an operating time;

[0013] S104, performing feature dimensionality reduction on the operating parameter sets of samples in the training set and the test set, retaining only the key parameter set that can affect all target parameters for each sample, and removing the remaining operating parameters from the sample.

[0014] Preferably, in step S104, after feature dimension reduction, the input of each sample is a set of key operating parameters of the unit including non-adjustable operating parameters and adjustable operating parameters, expressed as N is the total number of samples in the training set, p is the total number of non-adjustable operating parameters in the key parameter set, and q is the total number of adjustable operating parameters in the key parameter set; the output of the training set is all boiler combustion efficiency indicators and pollutant emission indicators that need to be optimized, expressed as the target parameter set m is the target parameter to be optimized The total number of , i = 1, 2, ..., N, j = 1, 2, ..., m; the boiler combustion efficiency index adopts an index form that is negatively correlated with the boiler combustion efficiency. The smaller the index value, the higher the corresponding boiler combustion efficiency. The pollutant emission index adopts an index form that is positively correlated with the pollutant emission. The smaller the index value, the smaller the corresponding pollutant emission.

[0015] Preferably, in step S2, it specifically includes the following steps:

[0016] S201. Normalize the sample data in the training set;

[0017] S202. For m target parameters, establish a multi-input single-output gradient descent decision tree model respectively to simulate the boiler combustion process. The input of each gradient descent decision tree model is p + q key parameters in the key parameter set, and the output is one target parameter Train each model based on the training set data to complete parameter optimization;

[0018] S203. After performing the same normalization process on the test set data as on the training set data, verify the accuracy of each gradient descent decision tree model that has completed parameter optimization. When the accuracy requirement is met, complete the model training for subsequent multi-objective optimization; otherwise, retrain the model.

[0019] Preferably, in step S202, the gradient descent decision tree model uses sampling without replacement. First, adjust the hyperparameters for the number of weak learners and the learning rate, and then further adjust other various hyperparameters of the weak learners.

[0020] Preferably, in step S3, it specifically includes the following steps:

[0021] S301. Determine the objective function and constraint conditions required for the multi-objective optimization of the coal-fired boiler to determine the optimization model; among them, for the boiler combustion efficiency index and pollutant emission index, an objective function for minimizing each index needs to be set, and the constraint conditions need to be set according to the allowable adjustment range of each parameter in the coal-fired boiler;

[0022] S302. Select the NSGA-II algorithm as the genetic algorithm for multi-objective optimization, and first set the parameters of the genetic algorithm; obtain the key parameter set at the current moment in real time from the operation process of the coal-fired boiler, and after performing the same normalization process on it as on the training set data, use it for population initialization;

[0023] S303. Based on the initial population, perform iterative optimization on each parameter in the key parameter set based on the NSGA-II algorithm. In each step of the iterative process of optimization, calculate the values of each objective function based on the key parameter set corresponding to the population individuals through the gradient descent decision tree model trained in S2 for population individual screening; after the iterative optimization is completed, obtain a set of Pareto optimal solution sets.

[0024] Preferably, in step S301, the established optimization model is as follows:

[0025]

[0026] Among them: The optimization objective is to make all boiler combustion efficiency indicators and pollutant emission indicators as small as possible. respectively represent the target parameters For the corresponding gradient descent decision tree model, in the input of each model, v represents the non-adjustable operating parameters in the set of key parameters, and x represents the adjustable operating parameters in the set of key parameters; g i (x) represents the adjustment amplitude of the i-th adjustable operating parameter, and the constraint condition g i (x) ≤ δ means that for the i-th adjustable operating parameter, its adjustment amplitude is restricted within δ above and below the current actual value.

[0027] Preferably, in step S302, the parameters of the genetic algorithm are set as follows: the population size is 100, the maximum number of generations is 100, the generation gap is 0.8, the crossover probability is 0.9, the mutation probability is 0.1, and the initial population P is randomly generated.

[0028] Preferably, in step S303, when using the NSGA-II algorithm for optimization, the initial population P generates offspring Q through simulated binary crossover and polynomial mutation within the range of the constraint conditions, combines P and Q into the population R for fast non-dominated sorting and crowding distance calculation, and selects individuals with better performance and the size of the population from R as the next parent generation according to the quality of the individuals, and repeats until the maximum number of generations is reached.

[0029] Preferably, in step S4, it specifically includes the following steps:

[0030] S401. Search in the Pareto optimal solution set obtained in S3 to see if there is a situation where all objective functions decrease. If so, preferentially select the optimization result where all objective functions decrease and the total decrease amount after adding the objective functions is the largest as the optimal solution. If not, select the optimization result with the largest total decrease amount after adding the objective functions as the optimal solution;

[0031] S402. According to the set of key parameters corresponding to the optimal solution selected in S401, reverse-normalize the adjustable operating parameters therein and output them for controlling the operation of the coal-fired boiler.

[0032] The present invention has the following beneficial effects compared with the prior art:

[0033] The present invention comprehensively considers multiple operating variables for multi-objective variable optimization, and can realize the optimization of multi-objective parameters related to improving boiler combustion efficiency and reducing pollutant emissions on the basis of establishing a boiler combustion model, so as to achieve the goal of improving combustion efficiency and reducing pollutant emissions. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a flowchart of a multi-objective optimization method for a coal-fired boiler based on an evolutionary algorithm.

[0035] Figure 2 It is the test result of the test set of the present invention on the decision tree model of the temperature gradient decline at the outlet of the A air preheater.

[0036] Figure 3 It is Figure 2 the test results of 300 samples among them.

[0037] Figure 4 It is the algorithm flowchart of the multi-objective optimization solution of the coal-fired boiler by evolving with the NSGA-II algorithm of the present invention.

[0038] Figure 5 It is the selection process of the optimal solution in the embodiment of the present invention. Specific embodiments

[0039] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will describe the specific embodiments of the present invention in detail with reference to the accompanying drawings. Many specific details are set forth in the following description in order to fully understand the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below. The technical features in the various embodiments of the present invention can be combined correspondingly without conflict.

[0040] In a preferred embodiment of the present invention, a multi-objective optimization method for a coal-fired boiler based on an evolutionary algorithm is provided, which includes the following steps:

[0041] S1. Collect historical data of the operating parameter set and the target parameter set during the operation of the coal-fired boiler, where the operating parameter set includes two types: non-adjustable operating parameters and adjustable operating parameters; divide the historical data into a training set and a test set, and perform dimensionality reduction on the operating parameter sets included in the samples in the training set and the test set, and retain the key parameter set that can affect the target parameters.

[0042] In the embodiment of the present invention, the above step S1 specifically includes the following steps:

[0043] S101. According to the preset operating parameter set and target parameter set, collect historical data of each operating parameter and target parameter during the operation of the coal-fired boiler; the target parameter set includes the boiler combustion efficiency index and the pollutant emission index, and the operating parameter set includes all non-adjustable operating parameters and adjustable operating parameters related to any target parameter in the coal-fired boiler;

[0044] S102, preprocessing the historical data, removing outliers, and then dividing the historical data into a training set and a test set, wherein each sample in the training set and the test set includes an operating parameter set and a target parameter set at an operating time;

[0045] S104, performing feature dimensionality reduction on the operating parameter sets of samples in the training set and the test set, retaining only the key parameter set that can affect all target parameters for each sample, and removing the remaining operating parameters from the sample.

[0046] In the embodiment of the present invention, in the above step S104, after feature dimension reduction, the input of each sample is a set of key operating parameters of the unit including non-adjustable operating parameters and adjustable operating parameters, which is expressed as N is the total number of samples in the training set, p is the total number of non-adjustable operating parameters in the key parameter set, and q is the total number of adjustable operating parameters in the key parameter set; the output of the training set is all boiler combustion efficiency indicators and pollutant emission indicators that need to be optimized, expressed as the target parameter set m is the target parameter to be optimized The total number of , i = 1, 2, ..., N, j = 1, 2, ..., m; the boiler combustion efficiency index adopts an index form that is negatively correlated with the boiler combustion efficiency. The smaller the index value, the higher the corresponding boiler combustion efficiency. The pollutant emission index adopts an index form that is positively correlated with the pollutant emission. The smaller the index value, the smaller the corresponding pollutant emission.

[0047] It should be noted that the parameters included in the operating parameter set and the target parameter set when collecting historical data in the present invention need to be determined according to the actual situation of the boiler and the target requirements to be optimized. The operating parameters in the operating parameter set should include collectable data indicators as much as possible to avoid missing key parameters. When the operating parameter set is subjected to feature dimension reduction, it can be screened in combination with correlation analysis and the physical meaning of the parameter variables themselves. In addition, for the target parameter set, it needs to be determined according to the boiler performance to be optimized. In this embodiment, it is set as the boiler combustion efficiency index and the pollutant emission index, and it is expected to improve the boiler combustion efficiency and reduce pollutant emissions after optimization. Since the boiler combustion efficiency is as high as possible, and multi-objective optimization generally needs to adopt the form of minimizing the objective function, it is actually necessary to screen the index or take negative values, so that the smaller the index value of the boiler combustion efficiency index is, the higher the corresponding boiler combustion efficiency is. For example, the air preheater outlet temperature is used in the subsequent embodiments. The air preheater outlet temperature represents the higher the boiler combustion efficiency.

[0048] S2. For each target parameter in the set of target parameters, train a boiler combustion process model using the training set and verify it using the test set. The boiler combustion process model in this embodiment is a gradient descent decision tree model that takes the set of key parameters as input and a single target parameter as output.

[0049] In an embodiment of the present invention, step S2 specifically includes the following steps:

[0050] S201. Normalize the sample data in the training set;

[0051] S202. For m target parameters, establish a multi-input single-output gradient descent decision tree model respectively to simulate the boiler combustion process. The input of each gradient descent decision tree model is p + q key parameters in the set of key parameters, and the output is a target parameter Train each model based on the training set data to complete parameter optimization;

[0052] S203. After performing the same normalization process on the test set data as on the training set data, verify the accuracy of each gradient descent decision tree model that has completed parameter optimization. When the accuracy requirement is met, complete the model training for subsequent multi-objective optimization; otherwise, retrain the model.

[0053] In an embodiment of the present invention, in step S202 above, the gradient descent decision tree model uses sampling without replacement. First, adjust the hyperparameters for the number of weak learners and the learning rate, and then further adjust other various hyperparameters of the weak learners.

[0054] S3. Set the objective function and constraint conditions according to the set of target parameters of the coal-fired boiler. Initialize the population by obtaining the set of key parameters at the current moment in real time during the operation of the coal-fired boiler. Iteratively optimize the key parameters in the set of key parameters through the NSGA-II algorithm. In each iteration process, calculate the objective function value corresponding to the population individuals through the boiler combustion process model trained in S2. After the optimization is completed, obtain the Pareto optimal solution set.

[0055] In an embodiment of the present invention, step S3 specifically includes the following steps:

[0056] S301. Determine the objective function and constraint conditions required for the multi-objective optimization of the coal-fired boiler, so as to determine the optimization model; among them, for the boiler combustion efficiency index and the pollutant emission index, an objective function for minimizing each index needs to be set, and the constraint conditions need to be set according to the allowable adjustment range of each parameter in the coal-fired boiler;

[0057] S302. Select the NSGA-II algorithm as the genetic algorithm for multi-objective optimization, and first set the parameters of the genetic algorithm; obtain the set of key parameters at the current moment in real time from the operation process of the coal-fired boiler, and after performing the same normalization processing as the training set data, use it for population initialization;

[0058] S303. Based on the initial population, perform iterative optimization on each parameter in the set of key parameters using the NSGA-II algorithm. In each step of the iterative optimization process, the gradient descent decision tree model trained in S2 is used to calculate the values of each objective function based on the set of key parameters corresponding to the population individuals (for each objective function, the gradient descent decision tree model corresponding to the target parameter needs to be called for calculation), which is used for population individual screening; after the iterative optimization is completed, a set of Pareto optimal solution sets is obtained.

[0059] In the embodiment of the present invention, in the above step S301, the established optimization model is as follows:

[0060]

[0061] Among them: the optimization goal is to make all the boiler combustion efficiency indicators and pollutant emission indicators reach as small as possible. respectively represent the target parameters corresponding gradient descent decision tree models. In the input of each model, v represents the non-adjustable operating parameters in the set of key parameters, and x represents the adjustable operating parameters in the set of key parameters; g i (x) represents the adjustment amplitude of the i-th adjustable operating parameter, and the constraint condition g i (x) ≤ δ means that for the i-th adjustable operating parameter, its adjustment amplitude is limited within δ above and below the current actual value. δ can be adjusted according to the actual situation, and is preferably 5% in this embodiment.

[0062] In the embodiment of the present invention, in the above step S302, the parameters of the genetic algorithm are set as follows: the population size is 100, the maximum number of generations of evolution is 100, the generation gap is 0.8, the crossover probability is 0.9, the mutation probability is 0.1, and the initial population P is randomly generated.

[0063] In the embodiment of the present invention, in the above step S303, when using the NSGA-II algorithm for optimization, the initial population P generates offspring Q through simulated binary crossover and polynomial mutation within the constraint range, combines P and Q into the population R for fast non-dominated sorting and crowding distance calculation, and selects individuals with better performance and the size of the population as the next parent from R, and repeats until the maximum number of generations of evolution is reached.

[0064] S4. Select again from the Pareto optimal solution set obtained in S3 to obtain an optimal solution with the highest degree of optimization, and adjust the operating parameters of the coal-fired boiler according to this optimal solution.

[0065] In the embodiment of the present invention, the above step S4 specifically includes the following steps:

[0066] S401. Check whether there is a situation where all objective functions decrease in the Pareto optimal solution set obtained in S3. If so, preferentially select the optimization result where all objective functions decrease and the total decrease amount after adding the objective functions is the largest as the optimal solution. If not, select the optimization result with the largest total decrease amount after adding the objective functions as the optimal solution.

[0067] It should be noted that since the parameters in the present invention are normalized in advance, there is comparability between different objective functions. By directly adding all objective function values and comparing them with the original values, the total decrease amount can be judged.

[0068] S402. According to the set of key parameters corresponding to the optimal solution selected in S401, output the adjustable operating parameters after inverse normalization for controlling the operation of the coal-fired boiler.

[0069] It should be noted that the inverse normalization here needs to be completely inverse to the normalization process in S302.

[0070] Next, the multi-objective optimization method for coal-fired boilers based on the evolutionary algorithm shown in S1 - S4 above will be applied to a real case of boiler combustion optimization in a thermal power plant to illustrate the specific implementation steps of the present invention and verify the effectiveness of the proposed method.

[0071] Embodiment

[0072] Refer to Figure 1 , which shows the flowchart of the multi-objective optimization method for coal-fired boilers based on the evolutionary algorithm in this embodiment. The method includes the following steps:

[0073] Step 1. Collect historical operation data during the boiler operation process, divide the training data set and the test data set, preprocess the data in the training data set, and select key parameters. Specifically, it includes the following steps:

[0074] 1.1) Select the unit operation parameter variables related to the boiler combustion efficiency and pollutant emissions;

[0075] 1.2) Extract the historical data of the unit operation parameters;

[0076] 1.3) Eliminate the outlier data and construct the training set and the test set;

[0077] 1.4) Combine correlation analysis and the physical meaning of variables to perform feature dimensionality reduction on the training set and the test set. After dimensionality reduction, only the key parameters that have a crucial impact on the target parameters are retained in the set of operating parameters, while the remaining non-critical parameters are eliminated. The input of a single sample in the training set after dimensionality reduction is a set of unit operating parameters including non-adjustable operating parameters and adjustable operating parameters, denoted as where N is the number of sample points in the training set, p is the total number of non-adjustable operating parameters, and q is the total number of adjustable operating parameters. The output of the training set is the set of performance indicators to be optimized, i.e., the target parameter set, denoted as where m is the total number of target parameters to be optimized.

[0078] It should be noted that there are two air preheaters in the coal-fired boiler of this embodiment, which are symmetrically arranged in the tail flue of the boiler. Therefore, the two different air preheaters are referred to as Air Preheater A and Air Preheater B. The tail gas of the boiler is input into the SCR denitration reactor, and the reactor is used to remove nitrogen oxides in the flue gas generated by the boiler combustion. There are two measuring points arranged in the reactor. Therefore, the names of these two measuring points are recorded as Reactor 1A and Reactor 1B.

[0079] In this embodiment, according to Step 1, after data preprocessing, the set of key parameters finally obtained by dimensionality reduction in the samples of the training set and the test set as the input contains 35 operating parameters (including 19 non-adjustable operating variables and 16 adjustable operating variables). The specific names are shown in Table 1, and the output is 4 performance indicators (namely, the outlet temperature of Air Preheater A, the outlet temperature of Air Preheater B, the NOx content in the inlet flue of Reactor 1A, and the NOx content in the inlet flue of Reactor 1B). The sampling frequency of the parameter variables in the following set of key parameters is 1 minute.

[0080] Table 1 Set of Key Parameters

[0081]

[0082]

[0083] It should be noted that among the above operating variable names, those marked as 1A, 1B, etc. refer to different positions of the same device (marked in italics); those marked as 1A4, 1B6, etc. refer to different devices (marked in bold and underlined, such as different burners).

[0084] Step 2: Establish a boiler combustion process model based on the training data set and verify it with the test data set. The combustion process model is a gradient descent decision tree model related to each target parameter. Specifically, it includes the following steps:

[0085] 2.1) Normalize the training set data.

[0086] 2.2) Based on the training set data, for m target parameters, establish multi-input single-output gradient descent decision tree models respectively, and adjust the model parameters.

[0087] In this embodiment, the gradient descent decision tree model uses sampling without replacement. The model needs to first adjust the hyperparameters and then perform model training. When adjusting the hyperparameters, in order to improve the adjustment efficiency, the hyperparameters of the number of weak learners and the learning rate can be adjusted first, and then other various hyperparameters of the weak learners can be further adjusted.

[0088] 2.3) Perform the same normalization process on the test set data as on the training set data, and verify the accuracy of the gradient descent decision tree model.

[0089] In this embodiment, according to step 2, the original data is divided into a training set and a test set according to a ratio of 2:1. Establish a gradient descent decision tree model for the outlet temperature of the A air preheater, a gradient descent decision tree model for the outlet temperature of the B air preheater, a gradient descent decision tree model for the NOx content in the inlet flue of reactor 1A, and a gradient descent decision tree model for the NOx content in the inlet flue of reactor 1B in the training set. Test the established models on the test set to verify the generalization ability of the established models. Refer to Figure 2 For the verification test results of the overall 2343 test samples in the test set on the gradient descent decision tree model of the outlet temperature of the A air preheater, Figure 3 For the verification test results of 300 of these samples. It can be seen from the figure that the predicted values and the true values are relatively close, so it proves that the accuracy of the model is relatively high.

[0090] Step 3: Select the NSGA-II algorithm as the constrained multi-objective optimization algorithm in this embodiment, determine the objective function and the constraint conditions, input the test samples, and perform iterative optimization on the operating variables based on the NSGA-II algorithm in the constrained multi-objective optimization algorithm to obtain the corresponding Pareto optimal solution set for each sample. Specifically, it includes the following steps:

[0091] 3.1) Determine the optimization model, mainly to determine the objective function and the constraint conditions;

[0092] The optimization model established in this embodiment is as follows:

[0093]

[0094] Among them, the optimization goal is to make both the outlet flue gas temperature and the NOx content as small as possible. v represents the non-adjustable operating parameters, v ∈ [generator power, coal feedback, secondary air damper, inlet air damper, oxygen content in the outlet flue, etc.], x represents the adjustable operating parameters, x ∈ [secondary air volume, secondary air damper, baffle valve position, burnout air damper, etc.], and the constraint condition g i(x) ≤ 0.05 means that for each operating variable, its adjustment range is limited within 5% above and below the current value.

[0095] 3.2) Set the parameters of the genetic algorithm and initialize the population for each test sample.

[0096] In this embodiment, the population size of the genetic algorithm is set to 100, the maximum number of generations is 100, the generation gap is 0.8, the crossover probability is 0.9, the mutation probability is 0.1, and the initial population P is randomly generated.

[0097] 3.3) Iteratively optimize the operating variables based on the NSGA-II algorithm to obtain a set of optimal solution sets corresponding to each test sample.

[0098] In this embodiment, the NSGA-II algorithm adopted belongs to the prior art. During the optimization process using the NSGA-II algorithm, the initial population P generates offspring Q through simulated binary crossover and polynomial mutation within the range of constraint conditions. P and Q are merged into population R for fast non-dominated sorting and crowding distance calculation. Individuals with better performance and the size of the population are selected from R as the next parent generation, and the process is repeated until the maximum number of generations is reached. The algorithm flow chart for solving using the NSGA-II algorithm is referred to Figure 4 .

[0099] Step 4: Select again from the optimized Pareto optimal solution set to obtain the solution with the highest degree of optimization. Specifically, it includes the following steps:

[0100] 4.1) Check whether there is a situation where all objective functions decrease in the obtained Pareto optimal solution set;

[0101] 4.2) First, select the optimization result where all objective functions decrease and decrease the most. Secondly, select the optimization result where the sum of the objective functions decreases the most. The output result is the optimization result of the current test sample and the corresponding adjustable operating parameter combination. It should be noted that since the key parameters output are normalized, inverse normalization should be performed correspondingly when actually used for control.

[0102] In this embodiment, according to Step 4, for the test sample, after iterative optimization, a set of optimal solution sets containing 100 solutions is obtained. Based on the objective function values of these solutions, a secondary selection is made to determine the optimal solution with the highest degree of optimization. The selection process of the optimal solution can be referred to Figure 5 , where the screening criterion for the highest degree of optimization can be executed according to the following criteria: First, select the optimization result where all 4 objective functions decrease and decrease the most; if there is no optimization result where all 4 objective functions decrease simultaneously in the optimal solution set, then select the optimization result where the sum of the 4 objective functions decreases the most.

[0103] Statistics on the optimization results were carried out for a total of 300 test samples: the average optimization rate of a single objective function can reach about 95%, and the optimization rate of the simultaneous reduction of the four objective functions can reach more than 89%; by adopting the optimized combination of operating variables, the average outlet temperature of the air preheater can be reduced by about 1.5 °C, and the average NO content in the inlet flue can be reduced by about 10 mg / 3 . It can be seen that the present invention has a high optimization rate in this case, and can achieve the goal of online adjusting the operating variables according to the boiler operation data to improve the combustion efficiency and reduce pollutant emissions.

[0104] The embodiments described above are only a preferred solution of the present invention, but it is not intended to limit the present invention. Those of ordinary skill in the relevant technical field can also make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, all technical solutions obtained by adopting the equivalent replacement or equivalent transformation method fall within the protection scope of the present invention.

Claims

1. A multi-objective optimization method for coal-fired boilers based on an evolutionary algorithm, characterized in that, The steps include: S1. Collect historical data of an operating parameter set and a target parameter set during the operation of a coal-fired boiler, wherein the operating parameter set includes two categories: non-adjustable operating parameters and adjustable operating parameters; divide the historical data into a training set and a test set, reduce the dimension of the operating parameter set contained in the samples in the training set and the test set, and retain the key parameter set that can affect the target parameter; S2. For each target parameter in the target parameter set, a boiler combustion process model is trained using the training set, and verified using the test set; the boiler combustion process model is a gradient descent decision tree model with the key parameter set as input and a single target parameter as output; S3. Setting the objective function and constraint conditions according to the target parameter set of the coal-fired boiler, thereby determining the optimization model; The key parameter set at the current moment is obtained in real time during the operation of the coal-fired boiler to initialize the population. The key parameters in the key parameter set are iteratively optimized through the NSGA-Ⅱ algorithm. In each iterative process, the objective function value corresponding to the population individual is calculated through the boiler combustion process model trained in S2. After the optimization is completed, the Pareto optimal solution set is obtained. The established optimization model is as follows: ； Among them: The optimization goal is to make all boiler combustion efficiency indicators and pollutant emission indicators as small as possible. respectively represent the target parameters For the corresponding gradient descent decision tree models, in the input of each model represents the non-adjustable operating parameters in the set of key parameters, represents the adjustable operating parameters in the set of key parameters; represents the adjustment range of the i-th adjustable operating parameter, and the constraint condition represents that for the i-th adjustable operating parameter, its adjustment range is restricted within the following and above the current actual value; S4. The Pareto optimal solution obtained in S3 is selected again to obtain an optimal solution with the highest degree of optimization, and the operating parameters of the coal-fired boiler are adjusted according to the optimal solution.

2. The multi-objective optimization method for a coal-fired boiler based on an evolutionary algorithm according to claim 1, wherein In step S1, the following steps are specifically included: S101. According to a preset operating parameter set and a target parameter set, historical data of each operating parameter and target parameter in the operation process of the coal-fired boiler are collected; the target parameter set includes a boiler combustion efficiency index and a pollutant emission index, and the operating parameter set includes all non-adjustable operating parameters and adjustable operating parameters related to any target parameter in the coal-fired boiler; S102, preprocessing the historical data, removing outliers, and then dividing the historical data into a training set and a test set, wherein each sample in the training set and the test set includes an operating parameter set and a target parameter set at an operating time; S104, performing feature dimensionality reduction on the operating parameter sets of samples in the training set and the test set, retaining only the key parameter set that can affect all target parameters for each sample, and removing the remaining operating parameters from the sample.

3. The multi-objective optimization method for a coal-fired boiler based on an evolutionary algorithm according to claim 2, characterized in that, In step S104, after feature dimensionality reduction, the input of each sample is a set of key unit operation parameters including non-adjustable operation parameters and adjustable operation parameters, expressed as , , where N is the total number of samples in the training set, is the total number of non-adjustable operation parameters in the set of key parameters, is the total number of adjustable operation parameters in the set of key parameters; the output of the training set is all the boiler combustion efficiency indicators and pollutant emission indicators that need to be optimized, expressed as the target parameter set , is the target parameter to be optimized of the total number, , ; among them, the boiler combustion efficiency indicator adopts an indicator form negatively correlated with the boiler combustion efficiency, and the smaller the indicator value, the higher the corresponding boiler combustion efficiency. The pollutant emission indicator adopts an indicator form positively correlated with the pollutant emission, and the smaller the indicator value, the less the corresponding pollutant emission.

4. The multi-objective optimization method for a coal-fired boiler based on an evolutionary algorithm according to claim 3, wherein, In step S2, the following steps are specifically included: S201, normalizing the sample data in the training set; S202. For target parameter, a gradient descent decision tree model with multiple inputs and a single output is established respectively to simulate the boiler combustion process. The input of each gradient descent decision tree model is key parameters in the key parameter set, and the output is a target parameter , and each model is trained based on the training set data to complete parameter optimization; S203, after normalizing the test set data in the same way as the training set data, the accuracy of each gradient descent decision tree model that has completed parameter optimization is verified. When the accuracy requirement is met, the model training is completed for subsequent multi-objective optimization, otherwise the model is retrained.

5. The multi-objective optimization method for coal-fired boilers based on an evolutionary algorithm according to claim 4, wherein In step S202, the gradient descent decision tree model adopts sampling without replacement, firstly adjusting the hyperparameters of the number of weak learners and the learning rate, and then further adjusting other hyperparameters of the weak learners.

6. The multi-objective optimization method for a coal-fired boiler based on an evolutionary algorithm according to claim 3, wherein In step S3, the following steps are specifically included: S301. Determine the objective functions and constraint conditions required for the multi-objective optimization of the coal-fired boiler, so as to determine the optimization model. Among them, for the boiler combustion efficiency index and pollutant emission index, an objective function for minimizing each index needs to be set, and the constraint conditions need to be set according to the allowable adjustment range of each parameter in the coal-fired boiler. S302. Select the NSGA-II algorithm as the genetic algorithm for multi-objective optimization, and first set the parameters of the genetic algorithm. Obtain the set of key parameters at the current moment in real time during the operation of the coal-fired boiler, and after performing the same normalization processing as the training set data, use it for population initialization. S303. Based on the initial population, perform iterative optimization on each parameter in the set of key parameters using the NSGA-II algorithm. In each step of the iterative optimization process, the gradient descent decision tree model trained in S2 is used to calculate the values of each objective function based on the set of key parameters corresponding to the population individuals for population individual screening. After the iterative optimization is completed, a set of Pareto optimal solution sets is obtained.

7. The multi-objective optimization method for a coal-fired boiler based on an evolutionary algorithm according to claim 6, wherein In step S302, the parameters of the genetic algorithm are set as follows: the population size is 100, the maximum number of generations is 100, the generation gap is 0.8, the crossover probability is 0.9, the mutation probability is 0.1, and the initial population is randomly generated .

8. The multi-objective optimization method for a coal-fired boiler based on an evolutionary algorithm according to claim 6, characterized in that In step S303, when using the NSGA-II algorithm for optimization, the initial population generates offspring through simulated binary crossover and polynomial mutation within the scope of the constraint conditions , and are merged into a population to perform fast non-dominated sorting and crowding distance calculation, and select individuals with better performance and the size of the population scale from as the next parental generation, repeating the process until the maximum number of evolutionary generations is reached.

9. The multi-objective optimization method for a coal-fired boiler based on an evolutionary algorithm according to claim 1, characterized in that In step S4, it specifically includes the following steps: S401. Search in the Pareto optimal solution set obtained in S3 to see if there is a situation where all objective functions decrease. If so, preferentially select the optimization result where all objective functions decrease and the total decrease amount after adding the objective functions is the largest as the optimal solution. If not, select the optimization result with the largest total decrease amount after adding the objective functions as the optimal solution. S402. According to the set of key parameters corresponding to the optimal solution selected in S401, re-denormalize the adjustable operating parameters among them and output them for controlling the operation of the coal-fired boiler.

Images