Generation method of optimal coal blending quality structure from multiple coal sources based on particle swarm optimization

Abstract

The present invention relates to a method for generating an optimal coal blending quality structure from multiple coal sources based on a particle swarm algorithm. Compared with the prior art, the method solves the problem that the quality of coal products required by different industries varies, resulting in differences in the utilization of coal and the changing laws of coal quality indicators, making it difficult to obtain a better coal blending quality structure. The present invention includes the following steps: acquisition of basic data from multiple coal sources and data preprocessing; additivity analysis of basic coal blending structure indicators; establishment of a coal blending quality prediction model; establishment of a nonlinear prediction model; construction of a multi-coal source coal blending quality structure optimization model; and generation of the optimal coal blending quality structure from multiple coal sources. The present invention uses a particle swarm algorithm in an adaptive form to establish a coal blending optimization model for coal industry indicators and conventional cohesiveness indicators, which can optimize the product structure and avoid resource waste.

Description

Technical Field

[0001] The invention relates to the technical field of coal industry analysis, in particular to a method for generating optimal coal blending quality structure of multiple coal sources based on a particle swarm algorithm. Background Art

[0002] There are few coal blending technologies for coal preparation plants. Due to the diversity of coal types, the products blended from multiple coal sources correspond to different industries. The changing patterns of the quality of raw coal after blending cannot be fully and accurately grasped, resulting in certain errors in the prediction of the quality of mixed coal, resulting in uneven coal blending effects.

[0003] The coal blending process is an important part of the production of coal preparation plants. The quality of coal blending directly affects the quality of subsequent products, and then affects the utilization of coal resources and the economic benefits of coal preparation plants. Conventional coal blending uses the weighted average method to estimate the analytical data of mixed coal. This method is feasible for coals with similar properties, but it is not feasible for coals with greatly different properties. Due to the different quality of coal products required by different industries, there are differences in the way coal is used and the change rules of coal quality indicators. For example, in the process of coking coal, the precipitation rule of mixed coal volatiles is affected by the type of coal blended.

[0004] Therefore, in view of the drawbacks of conventional coal blending methods such as the weighted average method, developing a coal blending model with optimal coal blending quality has become a technical problem that needs to be solved urgently. Summary of the invention

[0005] The purpose of the present invention is to solve the defect in the prior art that the quality of coal products required by different industries is different, which makes the utilization mode of coal and the change law of coal quality indicators different and it is difficult to obtain a better coal blending quality structure. A method for generating the optimal coal blending quality structure of multiple coal sources based on particle swarm algorithm is provided to solve the above problem.

[0006] In order to achieve the above object, the technical solution of the present invention is as follows:

[0007] The method for generating optimal coal blending quality structure of multiple coal sources based on particle swarm algorithm includes the following steps:

[0008] 11) Acquisition and preprocessing of basic data from multiple coal sources: Acquisition of single coal property data and generation of basic coal blending structure;

[0009] 12) Additivity analysis of basic coal blending structure indicators: Use the linear weighting method to determine whether the indicators of the basic coal blending structure are additivity;

[0010] 13) Establish a coal blending quality prediction model: Use SPSS software to analyze the correlation between coal blending quality indicators and different types of raw coal. For indicators that meet the additive property, use the multivariate linear regression method MLR to establish a coal blending quality prediction model;

[0011] 14) Establish a nonlinear prediction model: For indicators that do not meet the additivity, an indicator with a correlation greater than 0.7 for a single type of coal is introduced, and a nonlinear prediction model is established using a Gaussian kernel function;

[0012] 15) Constructing a multi-coal source coal blending quality structure optimization model: Taking the coal quality index requirements of coal blending as constraints and the amount of high-quality coal and low-quality coal as the objective function, a multi-coal source coal blending quality structure optimization model is constructed;

[0013] 16) Generation of optimal coal blending quality structure for multiple coal sources: The optimal coal blending quality structure is obtained using the particle swarm algorithm in an adaptive form.

[0014] The acquisition and data preprocessing of the multi-coal source basic data include the following steps:

[0015] 21) Collect experimental data for coal mining industry analysis, including coal source ratio information, single coal properties, single coal properties including ash content , moisture , volatile matter , sulfur content , adhesion index ;

[0016] 22) Set the coal blending ratio to ,in and ;

[0017] 23) The optimization method is used to design the mutual blending of two types of coal. That is, the test point is taken at the golden ratio division point of the interval, and the golden ratio division point is set to 0.168. The coal blending ratios of the two types of coal are obtained, and the relationship between the change of the mixed coal quality and the increasing proportion of a certain type of coal is obtained. The coal blending ratios of the three types of coal are obtained as the basic coal blending structure.

[0018] The additivity analysis of the basic coal blending structure index includes the following steps:

[0019] 31) Carry out coal blending experiments according to the basic coal blending structure, reduce and crush the clean coal products, obtain samples that meet the particle size requirements and obtain the ash content of each mixed coal , moisture , volatile matter , sulfur content , adhesion index ;

[0020] 32) According to the basic coal blending structure, the ash content of each coal type is linearly weighted in the case of two coals blending and three coals blending. , moisture , volatile matter , sulfur content , adhesion index , get the predicted values ​​of various indicators of mixed coal, and obtain the relationship between the coal quality characteristics of each single type of coal and its mixed coal,

[0021] The linear weighting formula is as follows:

[0022] ,

[0023] in, It indicates the coal quality characteristics of the corresponding mixed coal, that is, the weighted value of the coal quality index of each single coal. is the proportion of the i-th single type of coal, and , For the Coal quality index of a single type of coal, is the number of samples in the experimental coal blending;

[0024] 33) Compare the coal quality characteristics with the experimental data of coal mining industry analysis, and fit the experimental test data with the calculation results;

[0025] According to the comparison results, calculate the loss function in each fitting process ,in To calculate the fitted value, The actual result of the experiment is is the number of sample points in the fitting process;

[0026] 34) Using the linear calculation results obtained in step 32) and the degree of bias between the fitting curve and y=x, the ash content , moisture , volatile matter , sulfur content , adhesion index The additivity is judged: if the deviation of the graph drawn with the linear calculation result obtained in step 32) as the vertical axis and the actual coal quality data obtained in step 31) as the horizontal axis and the image of y=x is higher than the threshold, the condition of linear additivity is not met.

[0027] The establishment of the coal blending quality prediction model comprises the following steps:

[0028] 41) The ash obtained in step 31) , moisture , volatile matter , sulfur content , adhesion index The data was tested for abnormal data points using the Grubbs criterion proposed by Grubbs. The data after abnormal point detection and cleaning were imported into SPSS software in CSV file format and named train_coal. The Correlate module was selected for correlation analysis. Variables with correlation greater than 0.7 were selected to establish a multivariate linear regression quality prediction model.

[0029] 42) Select variables with correlation greater than 0.7 to establish a multivariate linear regression quality prediction model, that is, , moisture , volatile matter , sulfur content Establish a multivariate linear regression quality prediction model, and set the predicted value as On this basis, two parameters to be estimated are introduced ;

[0030] Establish a multivariate linear regression quality prediction model: ,

[0031] in, It represents the weighted value of coal quality index mixed with each single type of coal. Represents the mixed coal prediction index. In this model:

[0032] ;

[0033] in, Ash , moisture , volatile matter , sulfur content The predicted value of , , is the regression coefficient of the model, , , , is the regression intercept term of the model;

[0034] 43) Based on the multiple linear regression model and the cleaned data in step 41), import it into SPSS software, name it coal_regress, select the Regression module to perform multiple linear regression analysis and obtain the regression equation ;

[0035] 44) Perform statistical analysis on the obtained results to obtain evaluation indicators , where the regression sum of squares , the total sum of squares

[0036] The nonlinear prediction model is established, comprising the following steps:

[0037] 51) Establishing the bonding index Nonlinear prediction model ,

[0038] (5a) Volatile matter of fat coal Volatile matter of lean coal Perform nonlinear transformation to reduce the actual measurement Values ​​and predictions For the purpose of inter-value error, the Gaussian kernel function method is used to map finite-dimensional data into a high-dimensional space, which has the following form:

[0039] ,

[0040] in , Defined as a mapping from low dimension to high dimension, the kernel function represents the inner product after mapping to high-dimensional space, is the parameter to be estimated;

[0041] (5b) Add fat coal volatile matter and lean coal volatile matter The mixed term of Nonlinear prediction of values;

[0042] (5c) Import the model established in step (5b) and the cleaned data in step (41) into SPSS software and use the double iteration method to The solution is solved by using the quadratic programming algorithm established in each iteration to determine the solution direction, and then the parameters are calculated based on the minimum residual and SSE.

[0043] The construction of the multi-coal source coal blending quality structure optimization model comprises the following steps:

[0044] 61) The quality index prediction values ​​of the mixed coal obtained by the coal blending quality prediction model and the nonlinear prediction model are used to predict the quality index prediction values ​​including ash content. , moisture , volatile matter , sulfur content ; Sort to determine the first The lower limit of coal quality index is , the upper bound is ;

[0045] 62) Has A single coal, to prepare Quality indicators of mixed coal, A single type of coal was obtained through experiments The quality index is ,in ;

[0046] 63) Set The percentage of a single type of coal in the coal blend , then by The first type of coal The quality indicators are: , No. The lower limit of coal quality index is , the upper bound is , then use The first type of coal The coal quality index must be between the two.

[0047] Right now: ;

[0048] 64) Has Blending of single coals The single type of coal is high-quality coal, and its blending ratio is , then the objective function of minimizing the high-quality coal blending ratio is: ; Similarly, The single type of coal is low-quality coal, and the proportion of , then the objective function of minimizing the proportion of inferior coal is: ;

[0049] 65) The optimization model of coal quality structure of multiple coal sources is constructed as follows:

[0050]

[0051] In its constraints, the independent variable of the function They refer to the coal quality indexes of each mixed coal, among which the function , , , , They represent the prediction functions of the mixed bonding index, moisture, ash, total sulfur and volatile matter, respectively. For the The proportion of single coal.

[0052] The generation of the optimal coal blending quality structure of multiple coal sources comprises the following steps:

[0053] 71) The coal blending optimization process is represented in the form of real number coding. The position of the particle in the search space corresponds to the ratio of each single type of coal in the mixed coal scheme, and the dimension of the search space is the number of single types of coal added to the mixed coal. ,

[0054] The encoding method of real number encoding is as follows:

[0055] , is the type of influencing factors in the sample, that is, the first Optimization variables Map to Real numbers on ;

[0056] 72) Generate an initial group solution by random generation, that is, initialize the particle population;

[0057] 73) Define the position update method of individuals in the particle population, the expression is as follows:

[0058] ,

[0059] ,

[0060] In the formula, and Particles The historical best position vector of and the historical best position vector of the population, and Represents particles The velocity and position vectors, is the inertia factor during particle update, and For between A random value between , used to enhance the randomness of the search, and represents the learning factor;

[0061] 74) Determine the learning factor using particle swarm algorithm in adaptive form and , the calculation method is as follows:

[0062] ,

[0063] ,

[0064] in, and Respectively The maximum and minimum values ​​of ; Indicates the current particle swarm update search times; Indicates the maximum number of update searches;

[0065] 75) Determine the inertia factor using particle swarm algorithm in adaptive form , the calculation method is as follows:

[0066] ,

[0067] in, represents the maximum inertia weight, represents the minimum inertia weight, and , is the current iteration number;

[0068] 76) Repeatedly perform particle swarm optimization on the coal quality experimental data obtained in step 31), continuously update the search direction of the particle swarm and compare it with the dominant particles in the current search space, and select the better value to update the current optimal value; until the maximum number of iterations is reached or the global optimal position meets the limit; thereby obtaining the optimal coal blending quality structure.

[0069] Beneficial Effects

[0070] Compared with the prior art, the method for generating optimal coal blending quality structure of multiple coal sources based on particle swarm algorithm of the present invention uses particle swarm algorithm in an adaptive form to establish a coal blending optimization model for coal industry indicators and conventional cohesiveness indicators, which can optimize product structure and avoid waste of resources.

[0071] The present invention first uses the Grubbs criterion proposed by Grubbs to detect abnormal data points and perform data cleaning, which has strong scalability and excellent effect in processing complex data. The linear weighting method is used to determine whether the indicators are additive, and a nonlinear prediction model is established for indicators that are not additive. The Gauss kernel function is introduced to map finite-dimensional data to high-dimensional space, which improves the prediction effect of the model and enhances the generalization and robustness of the model.

[0072] The problem of solving the target parameters in the method for generating the optimal coal blending quality structure of multiple coal sources belongs to a multi-objective optimization problem. Commonly used intelligent optimization algorithms include simulated annealing algorithm, particle swarm algorithm, genetic algorithm, etc. Compared with other algorithms, the particle swarm algorithm in the adaptive form involved in the present invention has both the group intelligence background of the traditional evolutionary algorithm and good optimization performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 is a method sequence diagram of the present invention;

[0074] Figure 2 A framework diagram of the particle swarm algorithm (PSO) in an adaptive form involved in the present invention;

[0075] Figure 3 This is a fitness comparison diagram between the adaptive particle swarm algorithm involved in the present invention and the particle swarm algorithm. DETAILED DESCRIPTION

[0076] In order to have a further understanding and recognition of the structural features and the effects achieved by the present invention, a preferred embodiment and accompanying drawings are used for detailed description as follows:

[0077] like Figure 1 As shown, the method for generating the optimal coal blending quality structure of multiple coal sources based on the particle swarm algorithm of the present invention comprises the following steps:

[0078] The first step is to obtain and preprocess the basic data of multiple coal sources: obtain the property data of a single type of coal and generate the basic coal blending structure.

[0079] (1) Collect experimental data for coal mining industry analysis, including coal source ratio information and single coal properties, including ash content. , moisture , volatile matter , sulfur content , adhesion index .

[0080] (2) Set the coal blending ratio to ,in and .

[0081] (3) The optimization method is used to design the mutual blending of two types of coal. That is, the test point is taken at the golden ratio division point of the interval, and the golden ratio division point is set to 0.168. The coal blending ratios of the two types of coal are obtained, and the relationship between the change of the mixed coal quality and the increasing proportion of a certain type of coal is obtained. The coal blending ratios of the three types of coal are obtained as the basic coal blending structure.

[0082] The second step is to analyze the additivity of the basic coal blending structure indicators: use the linear weighting method to determine whether the indicators of the basic coal blending structure are additivity. The linear weighting method is used to determine whether the indicators are additivity, and a nonlinear prediction model is established for the indicators that are not additivity, which can achieve accurate prediction of various indicators in the model and enhance the reliability of model prediction.

[0083] (1) Carry out coal blending experiments according to the basic coal blending structure, reduce and crush the clean coal products, obtain samples that meet the particle size requirements and obtain the ash content of each mixed coal , moisture , volatile matter , sulfur content , adhesion index .

[0084] (2) According to the basic coal blending structure, the ash content of each coal type is linearly weighted in the case of two coals blended and three coals blended. , moisture , volatile matter , sulfur content , adhesion index , get the predicted values ​​of various indicators of mixed coal, and obtain the relationship between the coal quality characteristics of each single type of coal and its mixed coal,

[0085] The linear weighting formula is as follows:

[0086] ,

[0087] in, It indicates the coal quality characteristics of the corresponding mixed coal, that is, the weighted value of the coal quality index of each single coal. is the proportion of the i-th single type of coal, and , For the Coal quality index of a single type of coal, is the number of samples in the experimental coal blending.

[0088] (3) Compare the coal quality characteristics relationship with the experimental data of coal mining industry analysis, and fit the experimental test data with the calculation results;

[0089] According to the comparison results, calculate the loss function in each fitting process ,in To calculate the fitted value, The actual result of the experiment is is the number of sample points in the fitting process.

[0090] (4) Using the linear calculation results obtained in step (32) and the degree of bias between the fitting curve and y = x, the ash content , moisture , volatile matter , sulfur content , adhesion index The additivity is judged: if the deviation between the graph drawn with the linear calculation result obtained in step (2) as the vertical axis and the actual coal quality data obtained in step (1) as the horizontal axis and the image of y=x is higher than the threshold, the linear additivity condition is not met.

[0091] The third step is to establish a coal blending quality prediction model: use SPSS software to analyze the correlation between coal blending quality indicators and different types of raw coal. For indicators that meet the additivity requirement, use the multivariate linear regression method MLR to establish a coal blending quality prediction model.

[0092] In the quality prediction model, mixed coal indicators are often affected by multiple single coal quality indicators, and the multivariate linear regression method MLR can incorporate these factors into the model at the same time to more comprehensively analyze their combined impact on mixed coal indicators. At the same time, the model can also be improved based on new data and discoveries. If a certain input feature is found to have a collinearity problem with other input features, the input features can be processed through some methods such as principal component analysis PCA to improve the accuracy and stability of the model. This scalability and improvement enable the multivariate linear regression model to adapt to changing research needs and data conditions.

[0093] (1) The ash content obtained , moisture , volatile matter , sulfur content , adhesion index The Grubbs criterion proposed by Grubbs was used to detect abnormal data points. The data after abnormal point detection and cleaning were imported into SPSS software in CSV file format and named train_coal. The Correlate module was selected for correlation analysis, and variables with correlation greater than 0.7 were selected to establish a multivariate linear regression quality prediction model.

[0094] (2) Select variables with correlation greater than 0.7 to establish a multivariate linear regression quality prediction model, that is, , moisture , volatile matter , sulfur content Establish a multivariate linear regression quality prediction model, and set the predicted value as On this basis, two parameters to be estimated are introduced ;

[0095] Establish a multivariate linear regression quality prediction model: ,

[0096] in, It represents the weighted value of coal quality index mixed with each single type of coal. Represents the mixed coal prediction index. In this model:

[0097] ;

[0098] in, Ash , moisture , volatile matter , sulfur content The predicted value of , , is the regression coefficient of the model, , , , is the regression intercept term of the model.

[0099] (3) Based on the multiple linear regression model and the cleaned data, import them into SPSS software, name them coal_regress, select the Regression module to perform multiple linear regression analysis and obtain the regression equation .

[0100] (4) Perform statistical analysis on the obtained results to obtain evaluation indicators , where the regression sum of squares , the total sum of squares

[0101] The fourth step is to establish a nonlinear prediction model: For indicators that do not meet the additivity, the indicators with a correlation of a single type of coal greater than 0.7 are introduced, and a nonlinear prediction model is established using a Gaussian kernel function.

[0102] The Gaussian kernel function is a localized kernel function, which is characterized by having a larger weight near the data point and a rapidly decreasing weight far away from the data point, which means that it can capture the local characteristics of the data very well, so the nonlinear prediction model based on it usually has good generalization ability. At the same time, flexible parameter adjustment also makes it easy to combine with other machine learning methods to further improve the performance of the prediction model.

[0103] (1) Establishing the bonding index Nonlinear prediction model ,

[0104] (5a) Volatile matter of fat coal Volatile matter of lean coal Perform nonlinear transformation to reduce the actual measurement Values ​​and predictions For the purpose of inter-value error, the Gaussian kernel function method is used to map finite-dimensional data into a high-dimensional space, which has the following form:

[0105] ,

[0106] in , Defined as a mapping from low dimension to high dimension, the kernel function represents the inner product after mapping to high-dimensional space, is the parameter to be estimated;

[0107] (5b) Add fat coal volatile matter and lean coal volatile matter The mixed term of Nonlinear prediction of values;

[0108] (5c) Import the model established in step (5b) and the cleaned data in step (41) into SPSS software and use the double iteration method to The solution is solved by using the quadratic programming algorithm established in each iteration to determine the solution direction, and then the parameters are calculated based on the minimum residual and SSE.

[0109] The fifth step is to construct a quality structure optimization model for coal blending from multiple coal sources: the quality structure optimization model for coal blending from multiple coal sources is constructed with the coal quality index requirements for coal blending as constraints and the usage of high-quality coal and low-quality coal as the objective function.

[0110] The multi-coal source coal blending quality structure optimization model, which is constructed with the coal quality index requirements of coal blending as constraints and the amount of high-quality coal and low-quality coal as the objective function, has many technical advantages. It can accurately control the coal quality and optimize the amount of each coal source according to the set indicators to accurately blend coal, thereby improving combustion or production efficiency and quality. In terms of cost, it can reasonably adjust the ratio of high-quality and low-quality coal, reduce the cost of coal blending and improve the efficiency of coal resource utilization.

[0111] Compared with other coal blending methods, this model has significant advantages. Compared with the empirical coal blending method, it can meet the requirements of multiple coal quality indicators through precise calculations in terms of accuracy, avoiding quality fluctuations caused by subjective judgment or inaccurate grasp of complex relationships in empirical coal blending; in terms of cost control, it can comprehensively consider the coal source cost and coal quality indicators for comprehensive optimization, rather than just based on approximate price adjustments; in terms of adaptability, it can quickly generate new optimal solutions for changes in coal sources, unlike empirical coal blending, which requires re-exploration and uncertain results. Compared with the single-indicator adjustment coal blending method, it is more comprehensive and can simultaneously consider multiple coal quality indicators to meet the comprehensive requirements of multiple indicators such as environmental protection areas; it is more systematic, relying on clear objective functions and complete constraints to scientifically design and optimize coal blending plans, avoiding the problem of single-indicator methods losing sight of one thing while focusing on another.

[0112] (1) The quality index prediction values ​​of the mixed coal obtained by the coal blending quality prediction model and the nonlinear prediction model are used to predict the quality index prediction values ​​including ash content. , moisture , volatile matter , sulfur content ; Sort to determine the first The lower limit of coal quality index is , the upper bound is .

[0113] (2) If A single coal, to prepare Quality indicators of mixed coal, A single type of coal was obtained through experiments The quality index is ,in .

[0114] (3) Set The percentage of a single type of coal in the coal blend , then by The first type of coal The quality indicators are: , No. The lower limit of coal quality index is , the upper bound is , then use The first type of coal The coal quality index must be between the two.

[0115] Right now: .

[0116] (4) With Blending of single coals The single type of coal is high-quality coal, and its blending ratio is , then the objective function of minimizing the high-quality coal blending ratio is: ; Similarly, The single type of coal is low-quality coal, and the proportion of , then the objective function of minimizing the proportion of inferior coal is: .

[0117] (5) The optimization model of coal blending quality structure from multiple coal sources is constructed as follows:

[0118]

[0119] In its constraints, the independent variable of the function They refer to the coal quality indexes of each mixed coal, among which the function , , , , They represent the prediction functions of the mixed bonding index, moisture, ash, total sulfur and volatile matter, respectively. For the The proportion of single coal.

[0120] Step 6: Generation of the optimal coal blending quality structure for multiple coal sources: The optimal coal blending quality structure is obtained using the particle swarm algorithm under the adaptive form.

[0121] There are many significant technical benefits to using the particle swarm algorithm in an adaptive form to solve the optimal coal blending quality structure of multiple coal sources. Its global search capability is strong, and it can widely explore different coal source combinations in the solution space, spanning multiple possible solutions, effectively avoiding falling into local optimality, and accurately locating the global optimal solution that meets multiple coal quality indicators and optimizes the amount of high-quality and low-quality coal. Parameter adaptive adjustment gives the algorithm the ability to change flexibly during the search process. In the early stage, it relies on a large inertia weight to expand the global search range, and in the later stage, it timely transforms to focus on local search to improve accuracy. It is very comfortable in dealing with multi-constrained and multi-variable coal blending problems, and integrates coal quality indicator constraints into the fitness function screening scheme to achieve a good balance between search efficiency and solution accuracy. At the same time, the algorithm has good convergence, can quickly approach and accurately obtain the optimal solution, reduce computing resources and time consumption, and quickly respond to the need to adjust the coal blending scheme.

[0122] Compared with the traditional particle swarm algorithm, the particle swarm algorithm in the adaptive form shows unique advantages in solving the optimal coal blending quality structure of multiple coal sources. The inertia weight, learning factor and other parameters of the traditional particle swarm algorithm are usually fixed, which makes it difficult for the algorithm to balance the global search and local search capabilities during the search process. When dealing with the complex problem of coal blending of multiple coal sources, it is easy to converge to the local optimal solution too early, and it is impossible to fully explore the solution space to obtain the truly optimal coal blending solution. The particle swarm algorithm in the adaptive form can dynamically adjust these key parameters according to the algorithm iteration process. For example, a larger inertia weight is used in the early stage of the search to ensure that the particles widely explore the possibility of different coal source combinations. As the iteration progresses, the inertia weight is reasonably reduced and the learning factor is adjusted, so that the particles focus on local fine search, thereby more accurately meeting the requirements of multiple coal quality indicators and optimizing the amount of high-quality coal and low-quality coal. This adaptive feature allows the algorithm to not only converge faster in the multi-variable and multi-constrained scenario of coal blending of multiple coal sources, but also obtain a better coal blending quality structure, effectively improving the quality and efficiency of coal blending scheme design, and providing a more reliable technical guarantee for the efficient utilization of coal resources.

[0123] (1) The coal blending optimization process is represented in the form of real number coding. The position of the particle in the search space corresponds to the ratio of each type of coal in the coal blending scheme. The dimension of the search space is the number of single coals added to the coal blending. ,

[0124] The encoding method of real number encoding is as follows:

[0125] , is the type of influencing factors in the sample, that is, the first Optimization variables Map to Real numbers on .

[0126] (2) Generate an initial population solution by random generation, that is, initialize the particle population.

[0127] (3) Define the position update method of individuals in the particle population, which is expressed as follows:

[0128] ,

[0129] ,

[0130] In the formula, and Particles The historical best position vector of and the historical best position vector of the population, and Represents particles The velocity and position vectors, is the inertia factor during particle update, and For between A random value between , used to enhance the randomness of the search, and Represents the learning factor.

[0131] (4) Using the particle swarm algorithm in an adaptive form to determine the learning factor and , the calculation method is as follows:

[0132] ,

[0133] ,

[0134] in, and Respectively The maximum and minimum values ​​of ; Indicates the current particle swarm update search times; Indicates the maximum number of update searches.

[0135] (5) Determine the inertia factor using the particle swarm algorithm in an adaptive form , the calculation method is as follows:

[0136] ,

[0137] in, represents the maximum inertia weight, represents the minimum inertia weight, and , is the current iteration number.

[0138] (6) Repeatedly perform particle swarm optimization operations on the obtained coal quality experimental data, continuously update the search direction of the particle swarm and compare it with the dominant particles in the current search space, and select the better value to update the current optimal value; until the maximum number of iterations is reached or the global optimal position meets the limit; thereby obtaining the optimal coal blending quality structure.

[0139] like Figure 2 As shown in the figure, it is a flow chart of the method for generating the optimal coal blending quality structure of multiple coal sources based on the particle swarm algorithm involved in the present invention. In the optimization process of coal blending of multiple coal sources, the first thing to do is to obtain the basic data of multiple coal sources and fine data preprocessing. By performing additivity analysis on the basic coal blending structure indicators, the foundation is laid for subsequent accurate modeling. The coal blending quality prediction model and the nonlinear prediction model are established in turn to more accurately grasp the quality change law after mixing different coal sources. On this basis, a multi-coal source coal blending quality structure optimization model is constructed to provide a theoretical framework for realizing the optimal coal blending scheme. When solving the optimal coal blending quality structure of multiple coal sources, we introduced a powerful particle swarm algorithm. Starting from the initialization of the particle swarm, the individual adaptability of each particle is carefully calculated, and then the individual optimal position (Pbest) and the global optimal position (Gbest) are selected, and the particle speed and position are continuously updated. As the iteration proceeds, the current number of iterations and the inertia weight are updated, and it is continuously judged whether the global optimal position meets the limit and whether the maximum number of iterations is reached. Through this process, we can efficiently find the optimal quality structure of coal blending from multiple coal sources, achieve accurate allocation and efficient utilization of coal resources, meet the strict requirements of different industrial fields for coal quality, and at the same time promote the development of the coal industry towards intelligence, efficiency and greenness.

[0140] like Figure 3 As shown in the figure, it is a fitness comparison diagram of the adaptive morphological particle swarm algorithm and the traditional particle swarm algorithm in the method for generating the optimal coal quality structure of multiple coal sources based on the particle swarm algorithm involved in the present invention. It can be seen that the curve of the adaptive particle swarm algorithm is located above the particle swarm algorithm curve as a whole, and it can often obtain a higher fitness value. The fitness of the two algorithms has its own fluctuations at different sample numbers, but the fluctuation of the adaptive particle swarm algorithm is relatively small, and its fitness curve is smoother, indicating that the algorithm has better stability during the search process and can search more stably in the direction of a better solution. The fitness fluctuation of the particle swarm algorithm is relatively large, and it may be easier to fall into a local optimal solution or jump between different solutions during the search process, resulting in its unstable fitness value. In general, the adaptive particle swarm algorithm shows advantages in fitness, has better optimization performance and stability, and can more effectively find a better solution in this experiment or application scenario.

[0141] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions only describe the principles of the present invention. The present invention may be subject to various changes and improvements without departing from the spirit and scope of the present invention. These changes and improvements fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for generating optimal coal blending quality structure of multiple coal sources based on particle swarm algorithm, characterized in that: The following steps are involved: 11) Acquisition and preprocessing of basic data from multiple coal sources: Acquisition of single coal property data and generation of basic coal blending structure; 12) Additivity analysis of basic coal blending structure indicators: Use the linear weighting method to determine whether the indicators of the basic coal blending structure are additivity; 13) Establish a coal blending quality prediction model: Use SPSS software to analyze the correlation between coal blending quality indicators and different types of raw coal. For indicators that meet the additive property, use the multivariate linear regression method MLR to establish a coal blending quality prediction model; 14) Establish a nonlinear prediction model: For indicators that do not meet the additivity, an indicator with a correlation greater than 0.7 for a single type of coal is introduced, and a nonlinear prediction model is established using a Gaussian kernel function; 15) Constructing a multi-coal source coal blending quality structure optimization model: Taking the coal quality index requirements of coal blending as constraints and the amount of high-quality coal and low-quality coal as the objective function, a multi-coal source coal blending quality structure optimization model is constructed; The construction of the multi-coal source coal blending quality structure optimization model comprises the following steps: 151) The quality index prediction values ​​of the mixed coal obtained by the coal blending quality prediction model and the nonlinear prediction model are used to predict the quality index prediction values ​​including ash content. , moisture , volatile matter , sulfur content ; Sort to determine the first The lower limit of coal quality index is , the upper bound is ; 152) has A single coal, to prepare Quality indicators of mixed coal, A single type of coal was obtained through experiments The quality index is ,in ; 153) Set The percentage of a single type of coal in the coal blend , then by The first type of coal blending The quality indicators are: , No. The lower limit of coal quality index is , the upper bound is , then use The first type of coal blending The coal quality index must be between the two. Right now: ; 154) has If the single type of coal is mixed, The single type of coal is high-quality coal, and its blending ratio is , then the objective function of minimizing the high-quality coal blending ratio is: Similarly, The single type of coal is low-quality coal, and the proportion of , then the objective function of minimizing the proportion of inferior coal is: ; 155) The optimization model of coal quality structure of multiple coal sources is constructed as follows: ； In its constraints, the independent variable of the function They refer to the coal quality indexes of each mixed coal, among which the function , , , , They represent the prediction functions of the mixed bonding index, moisture, ash, total sulfur and volatile matter, respectively. For the The proportion of the single coals added; 16) Generation of optimal coal blending quality structure for multiple coal sources: The optimal coal blending quality structure is obtained using the particle swarm algorithm in an adaptive form.

2. The method for generating optimal coal blending quality structure of multiple coal sources based on particle swarm algorithm according to claim 1 is characterized in that: The acquisition and data preprocessing of the multi-coal source basic data include the following steps: 21) Collect experimental data for coal mining industry analysis, including coal source ratio information, single coal properties, single coal properties including ash content , moisture , volatile matter , sulfur content , adhesion index ; 22) Set the coal blending ratio to ,in and ; 23) The optimization method is used to design the mutual blending of two types of coal. That is, the test point is taken at the golden ratio division point of the interval, and the golden ratio division point is set to 0.

168. The coal blending ratios of the two types of coal are obtained, and the relationship between the change of the mixed coal quality and the increasing proportion of a certain type of coal is obtained. The coal blending ratios of the three types of coal are obtained as the basic coal blending structure.

3. The method for generating optimal coal blending quality structure of multiple coal sources based on particle swarm algorithm according to claim 2 is characterized in that: The additivity analysis of the basic coal blending structure index includes the following steps: 31) Carry out coal blending experiments according to the basic coal blending structure, reduce and crush the clean coal products, obtain samples that meet the particle size requirements and obtain the ash content of each mixed coal , moisture , volatile matter , sulfur content , adhesion index ; 32) According to the basic coal blending structure, the ash content of each coal type is linearly weighted in the case of two coals blending and three coals blending. , moisture , volatile matter , sulfur content , adhesion index , get the predicted values ​​of various indicators of mixed coal, and obtain the relationship between the coal quality characteristics of each single type of coal and its mixed coal, The linear weighting formula is as follows: ， in, It indicates the coal quality characteristics of the corresponding mixed coal, that is, the weighted value of the coal quality index of each single coal. is the proportion of the i-th single type of coal, and , For the Coal quality index of a single type of coal, is the number of samples in the experimental coal blending; 33) Compare the coal quality characteristics with the experimental data of coal mining industry analysis, and fit the experimental test data with the calculation results; According to the comparison results, calculate the loss function in each fitting process ,in To calculate the fitted values, The actual result of the experiment is is the number of sample points in the fitting process; 34) Using the linear calculation results obtained in step 32) and the degree of bias between the fitting curve and y=x, the ash content , moisture , volatile matter , sulfur content , adhesion index The additivity is judged: if the deviation of the graph drawn with the linear calculation result obtained in step 32) as the vertical axis and the actual coal quality data obtained in step 31) as the horizontal axis and the image of y=x is higher than the threshold, the condition of linear additivity is not met.

4. The method for generating optimal coal blending quality structure of multiple coal sources based on particle swarm algorithm according to claim 3 is characterized in that: The establishment of the coal blending quality prediction model comprises the following steps: 41) The ash obtained in step 31) , moisture , volatile matter , sulfur content , adhesion index The data was tested for abnormal data points using the Grubbs criterion proposed by Grubbs. The data after abnormal point detection and cleaning were imported into SPSS software in CSV file format and named train_coal. The Correlate module was selected for correlation analysis. Variables with correlation greater than 0.7 were selected to establish a multivariate linear regression quality prediction model. 42) Select variables with correlation greater than 0.7 to establish a multivariate linear regression quality prediction model, that is, , moisture , volatile matter , sulfur content Establish a multivariate linear regression quality prediction model, and set the predicted value as On this basis, two parameters to be estimated are introduced ; Establish a multivariate linear regression quality prediction model: , in, It represents the weighted value of coal quality index mixed with each single type of coal. Represents the mixed coal prediction index. In this model: ； in, Ash , moisture , volatile matter , sulfur content The predicted value of , , is the regression coefficient of the model, , , , is the regression intercept term of the model; 43) Based on the multiple linear regression quality prediction model and the cleaned data in step 41), import it into SPSS software, name it coal_regress, select the Regression module to perform multiple linear regression analysis and obtain the regression equation ; 44) Perform statistical analysis on the obtained results to obtain evaluation indicators , where the regression sum of squares , the total sum of squares 5. The method for generating optimal coal blending quality structure of multiple coal sources based on particle swarm algorithm according to claim 4 is characterized in that: The nonlinear prediction model is established, comprising the following steps: 51) Establishing the bonding index Nonlinear prediction model , (5a) Volatile matter of fat coal Volatile matter of lean coal Perform nonlinear transformation to reduce the actual measurement Values ​​and predictions For the purpose of inter-value error, the Gaussian kernel function method is used to map finite-dimensional data into a high-dimensional space, which has the following form: ， in , Defined as a mapping from low dimension to high dimension, the kernel function represents the inner product after mapping to high-dimensional space, is the parameter to be estimated; (5b) Add fat coal volatile matter and lean coal volatile matter The mixed term of Nonlinear prediction of values; (5c) Import the model established in step (5b) and the cleaned data in step (41) into SPSS software and use the double iteration method to The solution is solved by using the quadratic programming algorithm established in each iteration to determine the solution direction, and then the parameters are calculated based on the minimum residual and SSE.

6. The method for generating optimal coal blending quality structure of multiple coal sources based on particle swarm algorithm according to claim 5, characterized in that: The generation of the optimal coal blending quality structure of multiple coal sources comprises the following steps: 61) The coal blending optimization process is represented in the form of real number coding. The position of the particle in the search space corresponds to the ratio of each single type of coal in the mixed coal scheme, and the dimension of the search space is the number of single types of coal added to the mixed coal. , The encoding method of real number encoding is as follows: , is the type of influencing factors in the sample, that is, the first Optimization variables Map to Real numbers on ; 62) Generate an initial group solution by random generation, that is, initialize the particle population; 63) Define the position update method of individuals in the particle population, the expression is as follows: ， ， In the formula, and Particles The historical best position vector of and the historical best position vector of the population, and Represents particles The velocity and position vectors, is the inertia factor during particle update, and For between A random value between , used to enhance the randomness of the search, and represents the learning factor; 64) Determine the learning factor using particle swarm algorithm in adaptive form and , the calculation method is as follows: ， ， in, and Respectively The maximum and minimum values ​​of ; Indicates the current particle swarm update search times; Indicates the maximum number of update searches; 65) Determine the inertia factor using particle swarm algorithm in adaptive form , the calculation method is as follows: ， in, represents the maximum inertia weight, represents the minimum inertia weight, and , is the current iteration number; 66) Repeatedly perform particle swarm optimization on the coal quality experimental data obtained in step 31), continuously update the search direction of the particle swarm and compare it with the dominant particles in the current search space, and select the better value to update the current optimal value; until the maximum number of iterations is reached or the global optimal position meets the limit; thereby obtaining the optimal coal blending quality structure.

Images