Coal-fired boiler combustion optimization method and control method based on machine learning, system, equipment and medium

Abstract

The invention discloses a coal-fired boiler combustion optimization method, system and equipment based on machine learning and a medium, and belongs to the technical field of boiler operation optimizing.The method comprises the steps that operation parameter data and generated fly ash carbon content data are collected; constructing a machine learning model, and performing primary training on the machine learning model; the number of operation parameter data is reduced, secondary training is performed on the machine learning model, and the optimal parameter number is determined according to the fitting effect; screening the operation parameter data, and reserving the operation parameter data of which the quantity accords with the optimal parameter quantity; the method comprises the following steps: constructing a fly ash carbon content prediction model, optimizing an extreme value of the fly ash carbon content prediction model through an optimization algorithm, and determining an optimal boiler combustion operation parameter, and solves the problems that the actual measurement workload is large and the lowest fly ash carbon content cannot be obtained by only depending on a thermal-state multi-working-condition test and easily deviating from the optimal working condition; the existing prediction method has the problems that a local optimal solution is easy to generate and the generalization ability of a model is low.

Description

Technical Field

[0001] This invention relates to the field of boiler operation optimization technology, specifically to a combustion optimization method and control method, system, equipment and medium for coal-fired boilers based on machine learning. Background Technology

[0002] During boiler operation, the carbon content of fly ash is a key indicator that significantly impacts combustion efficiency. Incomplete combustion can also increase coal consumption for power generation and heating. Statistics show that for every 1% increase in boiler efficiency, coal consumption for power generation can decrease by approximately 1%, resulting in a substantial reduction in carbon emissions. Furthermore, the factors influencing fly ash carbon content are complex and diverse, including fixed parameters such as boiler design structure, operator skill levels, the properties of the coal fed into the boiler, and the combustion method. Therefore, establishing the relationship between fly ash carbon content and operating parameters, and precisely controlling and adjusting key parameters during boiler combustion to achieve optimized operation, is crucial. However, existing combustion optimization technologies face significant bottlenecks when dealing with complex operating conditions.

[0003] Most power plants in China optimize boiler operating parameters through hot-state multi-condition tests to achieve energy conservation, emission reduction, and lower consumption. However, the degree to which boiler operating parameters affect the carbon content of fly ash exhibits nonlinearity and strong coupling characteristics; relying solely on hot-state multi-condition tests involves a large workload and is prone to deviating from the optimal operating conditions, making it impossible to determine the lowest possible carbon content in fly ash.

[0004] Some studies have proposed optimizing and predicting fly ash carbon content based on principal component analysis (PCA) and a backpropagation (BP) neural network model, and applying a wolf pack algorithm to optimize boiler combustion parameters in multiple dimensions, significantly improving combustion efficiency. However, these prediction methods suffer from drawbacks such as the tendency to generate local optima and low model generalization ability. Summary of the Invention

[0005] In view of the above-mentioned problems, the present invention is proposed.

[0006] Therefore, the technical problem solved by this invention is that relying solely on hot multi-condition testing involves a large workload and is prone to deviating from the optimal operating conditions, making it impossible to obtain the lowest carbon content in fly ash; existing prediction methods have the drawbacks of easily generating local optima and having low model generalization ability.

[0007] To address the aforementioned technical problems, this invention provides the following technical solution: a machine learning-based method for optimizing and controlling the combustion of coal-fired boilers, comprising the following steps:

[0008] Conduct hot multi-condition tests, collect operating parameter data and fly ash carbon content data generated during the hot multi-condition tests, and perform preprocessing;

[0009] To build a machine learning model, the machine learning model is trained by taking the running parameter data as input and the generated fly ash carbon content data as output.

[0010] Reduce the amount of running parameter data, perform secondary training on the machine learning model, evaluate the fitting effect of the machine learning model according to the fitting evaluation index, and take the amount of running parameter data when the fitting effect is the best as the optimal number of running parameters.

[0011] Filter the operating parameter data and retain only those that meet the optimal number of operating parameters;

[0012] The machine learning model was optimized using the first optimization algorithm to construct a prediction model for the carbon content of fly ash.

[0013] The optimal boiler combustion operating parameters are determined by optimizing the fly ash carbon content prediction model using a second optimization algorithm.

[0014] As a preferred embodiment of the machine learning-based combustion optimization and control method for coal-fired boilers of the present invention, the preprocessing includes data scaling and data cleaning.

[0015] Data cleaning includes:

[0016] Set a data security threshold. When the operating parameter data and the generated fly ash carbon content data exceed the data security threshold, delete the corresponding operating parameter data and the generated fly ash carbon content data to obtain the missing part.

[0017] The missing parts were replaced using an outlier repair method.

[0018] As a preferred embodiment of the machine learning-based combustion optimization and control method for coal-fired boilers of the present invention, the step of evaluating the fitting effect of the machine learning model according to the fitting evaluation index includes:

[0019] Calculate the coefficient of determination between the predictions of the machine learning model and the actual values;

[0020] The corresponding machine learning models are selected by taking the maximum value of the coefficient of determination.

[0021] The number of parameter data for the selected machine learning models is set to the optimal number of parameters.

[0022] As a preferred embodiment of the machine learning-based combustion optimization and control method for coal-fired boilers of the present invention, the step of filtering operating parameter data includes:

[0023] Perform correlation analysis on the data of each operating parameter, obtain the correlation analysis results, and remove the corresponding operating parameter data based on the correlation analysis results;

[0024] The steps of correlation analysis include: quantifying the degree of linear correlation between operating parameter data, obtaining the correlation coefficient between operating parameter data, and removing the operating parameter data with the highest correlation coefficient.

[0025] The beneficial effects of this preferred technical solution are as follows:

[0026] By calculating the coefficient of determination and selecting the maximum value, the machine learning model with the best fitting effect can be accurately screened out. The number of data points corresponding to the selected machine learning model is set as the optimal number of operating parameters. Appropriately reducing the number of parameters can effectively avoid redundant calculations due to too many parameters or information loss due to too few parameters, thus providing a more scientific parameter basis for predicting the carbon content of fly ash.

[0027] As a preferred embodiment of the machine learning-based combustion optimization and control method for coal-fired boilers of the present invention, the optimization of the machine learning model through a first optimization algorithm includes:

[0028] The first optimization algorithm uses a swarm intelligence optimization algorithm;

[0029] Update the individual positions representing the parameters of the machine learning model according to the mathematical model and rules of the swarm intelligence optimization algorithm;

[0030] When the swarm intelligence optimization algorithm meets the preset convergence condition, it obtains the global optimal individual position and outputs the machine learning model parameters corresponding to the global optimal individual position as the optimal parameters of the machine learning model.

[0031] The beneficial effects of this preferred technical solution are as follows:

[0032] Swarm intelligence optimization algorithms possess strong global search capabilities. Using swarm intelligence optimization algorithms to optimize machine learning models can prevent them from getting trapped in local optima, making it easier to find the globally optimal parameters. Furthermore, the mathematical model and rules of swarm intelligence optimization algorithms are adapted to the individual position updates representing the machine learning model parameters, allowing for targeted adjustments to the model parameters and improving the accuracy of the machine learning model in predicting the carbon content of fly ash.

[0033] As a preferred embodiment of the machine learning-based coal-fired boiler combustion optimization and control method of the present invention, the step of optimizing the fly ash carbon content prediction model through a second optimization algorithm includes:

[0034] The extreme values ​​of fly ash carbon content predicted by the fly ash carbon content prediction model are optimized using the second optimization algorithm.

[0035] The minimum predicted carbon content in fly ash is taken as the optimization objective of the second optimization algorithm.

[0036] The optimal number of operating parameters is selected as the operating parameters to be optimized, and finally, the operating parameters to be optimized are filtered.

[0037] The beneficial effects of this preferred technical solution are as follows:

[0038] By optimizing the extreme values ​​of fly ash carbon content predicted by the fly ash carbon content prediction model using the second optimization algorithm, the operating parameters corresponding to the lowest fly ash carbon content can be accurately located, thereby improving the efficiency and accuracy of operating parameter optimization.

[0039] As a preferred embodiment of the machine learning-based combustion optimization and control method for coal-fired boilers of the present invention, the step of screening the operating parameters to be optimized includes:

[0040] Set the value range for the operating parameters to be optimized;

[0041] When the running parameter to be optimized exceeds the range of values, the corresponding running parameter to be optimized is discarded.

[0042] When the operating parameter to be optimized is within the specified range, the corresponding operating parameter to be optimized is taken as the optimal boiler combustion operating parameter.

[0043] This invention provides a combustion optimization system for coal-fired boilers based on machine learning.

[0044] To solve the above-mentioned technical problems, the present invention further provides the following technical solution: a coal-fired boiler combustion optimization system based on machine learning, comprising: a data acquisition module: acquiring operating parameter data and carbon content data of generated fly ash in hot multi-condition tests;

[0045] First data processing module: performs data scaling and data cleaning on operating parameter data and generated fly ash carbon content data;

[0046] The first model construction and prediction module: constructs a machine learning model and performs first-level training on the machine learning model;

[0047] The second data processing module performs secondary training on the machine learning model, evaluates the fitting effect of the machine learning model according to the fitting evaluation index, and takes the number of running parameter data when the fitting effect is the best as the optimal number of running parameters.

[0048] The third data processing module filters the operating parameter data based on the correlation analysis results, retaining only the number of operating parameter data that matches the optimal number of operating parameters.

[0049] The second model building module optimizes the machine learning model using the first optimization algorithm and builds a fly ash carbon content prediction model.

[0050] The fourth data processing module optimizes the fly ash carbon content prediction model using the second optimization algorithm to determine the optimal boiler combustion operation parameters.

[0051] The present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, characterized in that the processor, when executing the computer program, implements the steps of a machine learning-based combustion optimization method and control method for coal-fired boilers.

[0052] The present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the steps of a machine learning-based combustion optimization and control method for coal-fired boilers.

[0053] The beneficial effects of this invention are as follows: This invention establishes a parameter dataset of factors affecting fly ash carbon content through boiler hot-state multi-condition tests; it constructs a fly ash carbon content prediction model using machine learning methods, establishes the relationship between combustion parameters and fly ash carbon content, and optimizes the prediction model to improve its generalization ability; it optimizes the fly ash carbon content prediction model through an optimization algorithm to determine the optimal boiler combustion operating parameters. Furthermore, by using the optimization algorithm to optimize the extreme values ​​of fly ash carbon content predicted by the model, it can accurately locate the operating parameters corresponding to the lowest fly ash carbon content, improving the efficiency and accuracy of operating parameter optimization; the method proposed in this invention achieves global optimization of operating parameter data, reducing the generated fly ash carbon content and improving boiler combustion efficiency. Attached Figure Description

[0054] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 This is a flowchart illustrating a machine learning-based combustion optimization and control method for coal-fired boilers, as provided in one embodiment of the present invention. Detailed Implementation

[0056] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0057] Example 1, referring to Figure 1 This is the first embodiment of the present invention, which provides a combustion optimization and control method for coal-fired boilers based on machine learning, including steps S100 to S600:

[0058] S100: Collect and preprocess the operating parameter data and fly ash carbon content data generated during the hot multi-condition test.

[0059] S200. Input the operating parameter data into the machine learning model and output the fly ash carbon content data to perform first-level training on the machine learning model. Rank the operating parameter data by importance based on the results of the first-level training.

[0060] S300. By reducing the number of running parameter data, the machine learning model is trained in a second stage. The fitting effect of the machine learning model is evaluated according to the fitting evaluation index. The fitting effect is then filtered, and the number of running parameter data corresponding to the filtered fitting effect is taken as the optimal number of running parameters.

[0061] S400: Filter the operating parameter data and retain the operating parameter data whose quantity matches the optimal number of operating parameters;

[0062] S500: Optimize the machine learning model through the first optimization algorithm to construct a fly ash carbon content prediction model;

[0063] S600. The second optimization algorithm is used to optimize the extreme values ​​of the fly ash carbon content prediction model to determine the optimal boiler combustion operation parameters.

[0064] It should be noted that the combustion process of coal-fired boilers involves numerous operating parameters. The carbon content of fly ash, as a key indicator reflecting combustion efficiency, is influenced by a synergistic effect of multiple parameters, exhibiting a complex nonlinear relationship. Traditional combustion optimization methods struggle to accurately capture this nonlinear correlation, often relying on empirical adjustments, leading to low optimization efficiency and incomplete combustion. Furthermore, an excessive number of operating parameters increases model complexity, reducing optimization accuracy and speed. Different parameters have significantly different impacts on combustion efficiency, requiring targeted selection. In addition, a single model or algorithm cannot simultaneously achieve both prediction accuracy and optimization efficiency; a multi-level training and optimization strategy is necessary to accurately construct a fly ash carbon content prediction model and determine the optimal boiler combustion operating parameters. Data preprocessing, phased model training to optimize the number of parameters, and combining optimization algorithms to construct a prediction model and determine the optimal operating parameters are crucial for achieving efficient combustion and reducing energy consumption in coal-fired boilers.

[0065] Therefore, to address the aforementioned issues in operation monitoring and health prediction, a parameter dataset of factors affecting fly ash carbon content was established through steps S100-S600. A fly ash carbon content prediction model was constructed using machine learning methods to establish the relationship between combustion parameters and fly ash carbon content, and the prediction model was optimized to improve its generalization ability. An optimization algorithm was then used to further optimize the fly ash carbon content prediction model, determine the optimal boiler combustion operating parameters, achieve global optimization of operating parameter data, reduce the carbon content of generated fly ash, and improve boiler combustion efficiency.

[0066] Example 2, refer to Figure 1 This is the second embodiment of the present invention, which provides a combustion optimization method and control method for coal-fired boilers based on machine learning.

[0067] In this embodiment of the application, the preprocessing in step S100 includes data scaling and data cleaning:

[0068] Specifically, data scaling can be achieved through normalization, scaling the data to [0,1], as shown in the formula:

[0069]

[0070] Where, x new For the normalized data, x min x is the minimum value of the dataset. max This represents the maximum value in the dataset.

[0071] Data cleaning includes the following steps A1 to A2:

[0072] A1. Set a data security threshold. When the operating parameter data and the generated fly ash carbon content data exceed the data security threshold, delete the corresponding operating parameter data and the generated fly ash carbon content data to obtain the missing part.

[0073] A2. Replace the missing parts using outlier repair methods;

[0074] Specifically, in practical application, step A1 is as follows:

[0075] Set a maximum data security threshold and a minimum data security threshold. When the operating parameter data and the generated fly ash carbon content data are greater than the maximum data security threshold or less than the minimum data security threshold, delete the corresponding operating parameter data and the generated fly ash carbon content data to obtain the missing part.

[0076] Data security thresholds need to be determined comprehensively within a reasonable range, taking into account the physical meaning of the parameters, the safe operating conditions of the equipment, and the distribution of historical normal operating data.

[0077] Specifically, the outlier repair method in step A2 uses the boundary value replacement method. For data that is greater than the maximum data security threshold, it is directly replaced with the corresponding maximum data security threshold; for data that is less than the minimum data security threshold, it is directly replaced with the corresponding minimum data security threshold.

[0078] For missing values ​​in the collected operating parameter data and the generated fly ash carbon content data, linear interpolation was used to fill in the missing parts of the data.

[0079] In an optional implementation, the data scaling in step S100 can also employ a sigmoid function transformation to scale the data to [0,1], as shown in the following formula:

[0080]

[0081] Where x' is the data obtained after transformation by the sigmoid function, μ is the mean of the dataset, σ is the standard deviation of the dataset, and x is the data to be scaled.

[0082] In another optional implementation, the data scaling in step S100 can also employ a hyperbolic tangent function transformation to scale the data to [0,1], as shown in the following formula:

[0083]

[0084] Where x' is the data obtained after hyperbolic tangent function transformation, and x is the data to be scaled.

[0085] In this embodiment of the application, the machine learning model in step S200 can be a random forest model. The random forest model achieves accurate fitting of the carbon content of fly ash by integrating the prediction results of multiple decision trees. It can effectively handle the nonlinear correlation and strong coupling characteristics between high-dimensional operating parameters, and has strong robustness to outliers and noisy data.

[0086] The random forest model can quantify the contribution of each operating parameter data to the prediction result of fly ash carbon content, and rank the operating parameter data according to their contribution to the prediction result of fly ash carbon content, providing a basis for the subsequent screening of operating parameter data in S400.

[0087] In an optional implementation, the machine learning model in step S200 can also adopt a backpropagation neural network model. The backpropagation neural network model achieves accurate fitting of the carbon content of fly ash through multi-layer nonlinear mapping relationship, which can effectively capture the nonlinear correlation and strong coupling characteristics between high-dimensional operating parameters. Furthermore, by continuously adjusting the connection weights between neurons through the backpropagation algorithm, the modeling ability for complex combustion processes can be gradually optimized.

[0088] In another optional implementation, the machine learning model in step S200 can also adopt a multiple linear regression model, which achieves fitting by establishing a linear mapping relationship between multiple operating parameters and the carbon content of fly ash, and solves the regression coefficients by the least squares method to quantify the degree of linear influence of each operating parameter on the carbon content of fly ash.

[0089] In this embodiment of the application, step S300 evaluates the fitting effect of the machine learning model according to the fitting evaluation index, specifically: the coefficient of determination is used as the fitting evaluation index, and the machine learning model has the best fitting effect when the coefficient of determination is the largest.

[0090] The steps of filtering the fitting results and determining the optimal number of running parameters based on the number of data points corresponding to the filtered fitting results include B1 to B3:

[0091] B1. Calculate the coefficient of determination between the predicted values ​​and the actual values ​​of the machine learning model;

[0092] B2. Take the maximum value of the coefficient of determination and select the corresponding machine learning model;

[0093] B3. Set the number of running parameter data for the selected machine learning models to the optimal number of running parameters.

[0094] Specifically, the formula for calculating the coefficient of determination in step B1 is as follows:

[0095]

[0096] Among them, R 2 To determine the coefficient values, m is the number of samples used in training the machine learning model, and y i Let be the true value of the carbon content in the fly ash of the i-th sample. This represents the predicted carbon content of fly ash for the i-th sample by the machine learning model. This represents the average of the actual carbon content of fly ash.

[0097] By comparing the magnitudes of the determination coefficients calculated in step B1, the machine learning model with the largest determination coefficient is selected, and the number of running parameter data used by the machine learning model with the largest determination coefficient is set as the optimal number of running parameters.

[0098] In an alternative implementation, the fitting evaluation index can also be the mean absolute error, specifically formulated as follows:

[0099]

[0100] Where MAE is the mean absolute error, m is the number of samples in the validation set, and y i To verify the true value of the carbon content in fly ash of the i-th sample in the set. This is the predicted value of fly ash carbon content for the i-th sample by the machine learning model under the current parameter combination;

[0101] The smaller the mean absolute error, the better the fit of the current machine learning model.

[0102] In another alternative implementation, the root mean square error can also be used as the fitting evaluation index, with the specific formula as follows:

[0103]

[0104] Where RMSE is the root mean square error, m is the number of samples in the validation set, and y i To verify the true value of the carbon content in fly ash of the i-th sample in the set. This represents the predicted carbon content of fly ash for the i-th sample by the machine learning model under the current parameter combination.

[0105] The smaller the root mean square error, the better the fit of the current machine learning model.

[0106] In this embodiment of the application, step S400, which involves filtering the operating parameter data and retaining only the operating parameter data whose quantity matches the optimal number of operating parameters, specifically includes:

[0107] Based on the importance ranking of the running parameter data obtained in step S200, the running parameter data with the lowest importance in the importance ranking is eliminated one by one until the number of running parameter data meets the optimal number of running parameters.

[0108] In this embodiment of the application, the first optimization algorithm adopts a swarm intelligence optimization algorithm, and the optimization of the machine learning model by the first optimization algorithm in step S500 includes C1 to C2:

[0109] C1. Update the individual positions representing the parameters of the machine learning model according to the mathematical model and rules of the swarm intelligence optimization algorithm;

[0110] C2. When the swarm intelligence optimization algorithm meets the preset convergence condition, the global optimal individual position is obtained. At this time, the machine learning model parameters corresponding to the global optimal individual position are the optimal parameters of the machine learning model.

[0111] In this embodiment of the application, the swarm intelligence optimization algorithm can be the sparrow search algorithm;

[0112] In step C1, the parameters of the sparrow search algorithm are set as follows: the dimension of the parameters is set to 2, and the value range of each parameter is set, with the lower limit set to [1, 1] and the upper limit set to [1000, 500], thereby limiting the search space; the sparrow population size is set to 100, the maximum number of iterations is set to 200, the warning value is set to 0.7, the discoverer ratio is 0.4, and the scout ratio is 0.2.

[0113] In step C2, during the iterative process of the sparrow search algorithm, the fitness of each individual sparrow is continuously calculated. The algorithm continues until the number of iterations reaches the preset maximum of 200, or the change in the global optimal fitness of the population over 30 consecutive generations is less than 1 x 10^6. -5 When the convergence condition is met, the position vector of the globally optimal individual in the population at this time is analyzed to obtain the parameters of the machine learning model, which are the optimal parameters of the machine learning model.

[0114] Specifically, in the iterative process of the sparrow search algorithm, each set of sparrow positions represents a set of parameters for a machine learning model. In each iteration, the objective function value is calculated for the current parameter set, and five-fold cross-validation is used to evaluate the fitting effect of the machine learning model optimized by the sparrow search algorithm.

[0115] Specifically, the five-fold cross-validation method was used to evaluate the fitting effect of the machine learning model optimized by the sparrow search algorithm. This involved randomly and uniformly dividing the preprocessed running parameter data and the generated fly ash carbon content data into five mutually exclusive subsets.

[0116] One of the five subsets is selected as the validation set, and the remaining four are merged into the training set.

[0117] Train the machine learning model using the current training set;

[0118] The trained model is used to predict the carbon content of fly ash on the current validation set. The fitting evaluation effect is calculated by fitting evaluation index. The average value of the evaluation index obtained from 5 validations is taken as the fitting effect score of the current parameter combination to evaluate the fitting effect of the current parameter combination.

[0119] The sparrow search algorithm uses the fit score as the fitness function value to guide the sparrow population to update its position.

[0120] Specifically, if a certain combination of parameters has a better fit, the corresponding sparrow position is retained and used as the optimization direction for subsequent iterations until the sparrow search algorithm converges, thus obtaining the globally optimal combination of machine learning model parameters.

[0121] In an optional implementation, the first optimization algorithm in step C1 can also employ a flower pollination algorithm. The position vector of each pollen individual corresponds to a set of parameters of a machine learning model. The dimension of the position vector is equal to the number of parameters to be optimized. The value range of each dimension component is limited to a reasonable range of the corresponding parameters. A global search is performed with a probability p, usually set to 0.8, to simulate the process of pollen spreading across flowers by insects. A local search is performed with a probability of (1-p) to simulate the process of pollen self-pollination or pollination of neighboring flowers. After each iteration, the fitting evaluation index of the model parameter combination corresponding to each pollen individual under five-fold cross-validation is calculated as the fitness value of the pollen. When the number of iterations reaches a preset maximum value or the fitness value of the pollen does not significantly improve for 30 consecutive generations, the parameter combination corresponding to the globally optimal pollen position is the optimal parameter combination of the machine learning model.

[0122] In another optional implementation, the first optimization algorithm in step C1 can also adopt a particle swarm optimization algorithm. Each particle corresponds to a set of machine learning model parameters. The position vector of the particle represents the parameter combination, and the velocity vector represents the direction and magnitude of parameter adjustment. Both the position component and the velocity component are limited to a reasonable range. The particle update is guided by individual optimality and global optimality. After each iteration update, the fitting evaluation index of the parameter combination corresponding to each particle under five-fold cross-validation is calculated. If the current fitness is better than the particle's historical best, the historical best position of the corresponding particle is updated. If it is better than the global best, it is updated to the global best position of the entire population. When the number of iterations reaches a preset maximum value or the fitness value corresponding to the global best position is less than a set threshold for 50 consecutive generations, the parameter combination corresponding to the global best position is the optimal parameter combination of the machine learning model.

[0123] In the embodiments of this application, the machine learning model adopts the random forest model, and the parameter combination of the machine learning model is the number of decision trees and the depth of the decision trees in the random forest model.

[0124] In this embodiment of the application, the root mean square error can be used to calculate the fitting evaluation effect through the fitting evaluation index, and the specific formula is as follows:

[0125]

[0126] Where RMSE is the root mean square error, m is the number of samples in the validation set, and y i To verify the true value of the carbon content in fly ash of the i-th sample in the set. This represents the predicted carbon content of fly ash for the i-th sample by the machine learning model under the current parameter combination.

[0127] The average root mean square error obtained from the five validations is taken. The smaller the average root mean square error, the better the fitting effect of the parameter combination of the current machine learning model.

[0128] If the average value of the root mean square error of a certain parameter combination is smaller, the corresponding position of the sparrow is retained and used as the optimization direction for subsequent iterations.

[0129] In this embodiment of the application, step S600 uses a second optimization algorithm to optimize the extreme values ​​of the fly ash carbon content prediction model, determining the optimal boiler combustion operating parameters, including D1 to D2:

[0130] D1. The minimum predicted carbon content of fly ash is taken as the optimization target of the second optimization algorithm, and the data of the optimal number of operating parameters are taken as the operating parameters to be optimized.

[0131] D2. Filter the operating parameters to be optimized according to the set range to obtain the optimal boiler combustion operating parameters;

[0132] The second optimization algorithm used in step D1 is the improved sparrow search algorithm;

[0133] Specifically, the improved sparrow search algorithm is based on the sparrow search algorithm and incorporates Logistic-Tent chaotic mapping, Levy flight, and adaptive t-distribution mutation strategy.

[0134] The initialization process of SSA is optimized by using Logistic-Tent chaotic mapping, which effectively solves the problem of uneven random distribution of the initial position of sparrow population in traditional SSA, thereby increasing the possibility of escaping local optima.

[0135] The adaptive t-distribution mutation mechanism can adjust the degrees of freedom of the t-distribution according to the iteration process, thereby enhancing the diversity of the search process;

[0136] By embedding the Levy flight strategy in the algorithm, the position is updated with a certain probability based on the step size of the Levy flight, which improves the algorithm's ability to escape local optima and optimizes search performance.

[0137] The improved sparrow search algorithm, after the above optimizations, demonstrates excellent global extremum search capabilities.

[0138] In step D2, the operating parameters to be optimized are filtered according to the set range to obtain the optimal boiler combustion operating parameters:

[0139] Set the range of operating parameters to be optimized;

[0140] If the running parameter to be optimized is outside the range, discard the corresponding running parameter to be optimized.

[0141] When the operating parameter to be optimized is within the range, the corresponding operating parameter to be optimized is taken as the optimal boiler combustion operating parameter;

[0142] The corresponding mathematical model can be expressed as:

[0143]

[0144] Where F = f(x) is the fitness function of the improved sparrow search algorithm, x i For the i-th running parameter, x i,min The minimum value set for the i-th running parameter, x i,max The maximum value set for the i-th running parameter;

[0145] The specific range of values ​​for each operating parameter is determined based on actual operating experience and test data, taking into account factors such as equipment safety and process requirements.

[0146] Through the above steps, the operating parameter data corresponding to the minimum carbon content in fly ash is finally obtained.

[0147] Example 3 is the third embodiment of the present invention. This embodiment provides a coal-fired boiler combustion optimization system based on machine learning, including...

[0148] Data acquisition module: Collects operating parameter data and fly ash carbon content data during hot multi-condition tests;

[0149] First data processing module: performs data scaling and data cleaning on operating parameter data and generated fly ash carbon content data;

[0150] The first model construction and prediction module: constructs a machine learning model and performs first-level training on the machine learning model;

[0151] The second data processing module performs secondary training on the machine learning model, evaluates the fitting effect of the machine learning model according to the fitting evaluation index, and takes the number of running parameter data when the fitting effect is the best as the optimal number of running parameters.

[0152] The third data processing module filters the operating parameter data based on the correlation analysis results, retaining only the number of operating parameter data that matches the optimal number of operating parameters.

[0153] The second model building module optimizes the machine learning model using the first optimization algorithm and builds a fly ash carbon content prediction model.

[0154] The fourth data processing module optimizes the fly ash carbon content prediction model using the second optimization algorithm to determine the optimal boiler combustion operation parameters.

[0155] Example 4, the fourth embodiment of the present invention, differs from the previous three embodiments in that: if the function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0156] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0157] More specific examples (a non-exhaustive list) of computer-readable media include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which programs can be printed, because programs can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0158] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0159] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for optimizing combustion in a coal-fired boiler based on machine learning, characterized in that, include: Collect and preprocess operational parameter data and fly ash carbon content data from the hot multi-condition test. The operating parameter data is used as input to the machine learning model, and the fly ash carbon content data is used as output to perform first-level training on the machine learning model. The importance of the operating parameter data is ranked based on the results of the first-level training. By reducing the number of running parameter data, the machine learning model is trained in a second stage. The fitting effect of the machine learning model is evaluated according to the fitting evaluation index. The fitting effect is then filtered, and the number of running parameter data corresponding to the filtered fitting effect is taken as the optimal number of running parameters. Filter the operating parameter data and retain only those that meet the optimal number of operating parameters; The machine learning model is optimized using the first optimization algorithm to construct a fly ash carbon content prediction model. The optimal boiler combustion operating parameters are determined by optimizing the extreme values ​​of the fly ash carbon content prediction model using a second optimization algorithm.

2. The method for optimizing combustion in a coal-fired boiler based on machine learning as described in claim 1, characterized in that, The specific steps for filtering the runtime parameter data are as follows: Based on the importance of the operational parameter data, the operational parameter data with the lowest importance in the importance ranking are eliminated one by one until the number of operational parameter data meets the optimal number of operational parameters.

3. The method for optimizing combustion in a coal-fired boiler based on machine learning as described in claim 2, characterized in that, Optimizing the machine learning model using the first optimization algorithm includes: The first optimization algorithm employs a swarm intelligence optimization algorithm; Update the individual positions representing the parameters of the machine learning model according to the mathematical model and rules of the swarm intelligence optimization algorithm; When the swarm intelligence optimization algorithm meets the preset convergence condition, it obtains the global optimal individual position and outputs the machine learning model parameters corresponding to the global optimal individual position as the optimal parameters of the machine learning model.

4. The method for optimizing the combustion of coal-fired boilers based on machine learning as described in claim 3, characterized in that, The optimal boiler combustion operating parameters are determined by optimizing the extremum of the fly ash carbon content prediction model using a second optimization algorithm, including: The predicted carbon content of fly ash is taken as the minimum value as the optimization target of the second optimization algorithm, and the operating parameter data of the optimal number of operating parameters is taken as the operating parameters to be optimized. The optimal boiler combustion operating parameters are obtained by filtering the parameters to be optimized within a set range.

5. The method for optimizing the combustion of coal-fired boilers based on machine learning as described in claim 4, characterized in that, The steps for filtering the operating parameters to be optimized include: Set the range of operating parameters to be optimized; If the running parameter to be optimized is outside the range, discard the corresponding running parameter to be optimized. When the operating parameters to be optimized are within the specified range, the corresponding operating parameters to be optimized are taken as the optimal boiler combustion operating parameters.

6. The method for optimizing combustion in a coal-fired boiler based on machine learning as described in claim 5, characterized in that, The steps for evaluating the fitting performance of the machine learning model based on the fitting evaluation metrics include: The coefficient of determination is used as the fitting evaluation index. When the coefficient of determination is the largest, the machine learning model has the best fitting effect. Calculate the coefficient of determination between the prediction results of the machine learning model and the actual values; Filter out machine learning models whose coefficient of determination reaches its maximum value; Set the number of running parameter data for the selected machine learning model as the optimal number of running parameters.

7. The method for optimizing combustion in a coal-fired boiler based on machine learning as described in claim 6, characterized in that, The preprocessing includes data scaling and data cleaning; The data cleaning includes: A data security threshold is set. When the operating parameter data and fly ash carbon content data exceed the data security threshold, the corresponding operating parameter data and fly ash carbon content data are deleted to obtain the missing part. The missing parts were replaced using an outlier repair method.

8. A machine learning-based combustion optimization system for coal-fired boilers, employing the machine learning-based combustion optimization method for coal-fired boilers as described in any one of claims 1 to 7, characterized in that, include: Data acquisition module: Collects operating parameter data and fly ash carbon content data during hot multi-condition tests; First data processing module: performs data scaling and data cleaning on the operating parameter data and the fly ash carbon content data; First model construction and prediction module: Constructs a machine learning model and performs first-level training on the machine learning model; The second data processing module performs secondary training on the machine learning model, evaluates the fitting effect of the machine learning model according to the fitting evaluation index, and takes the number of running parameter data when the fitting effect is the best as the optimal number of running parameters. The third data processing module filters the operating parameter data based on the correlation analysis results, retaining only the number of operating parameter data that matches the optimal number of operating parameters. Second model construction module: Optimizes the machine learning model through the first optimization algorithm to construct a fly ash carbon content prediction model; The fourth data processing module optimizes the fly ash carbon content prediction model using the second optimization algorithm to determine the optimal boiler combustion operation parameters.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the machine learning-based combustion optimization method for coal-fired boilers as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the machine learning-based combustion optimization method for coal-fired boilers as described in any one of claims 1 to 7.

Images