Coking coal blending method based on intelligent optimization

Abstract

The invention relates to a coking coal blending method based on intelligent optimization, and the method comprises the following steps: S1, collecting historical feed coal mass ratio data and corresponding coke quality data, carrying out data cleaning preprocessing and data standardization, and analyzing the correlation between each coke quality parameter and a feed coal quality parameter; s2, performing hyper-parameter optimization of a base algorithm in combination with a correlation analysis result and historical data, determining each layer of base algorithm composition of a multi-algorithm multi-layer superposition model, and obtaining a coke parameter prediction model set; and S3, according to the coke quality constraint, on the basis of the coke parameter prediction model set, establishing a multi-constraint single-target random optimization model, and searching a lowest cost ratio scheme. According to the method, manual coal blending is converted into intelligent coal blending, the model is trained according to historical data, the proportioning scheme is automatically generated through the model, the manual participation degree is reduced, the process is simplified, and the proportioning cost and the proportioning time are lower.

Description

Technical Field

[0001] The invention belongs to the technical field of coal chemical industry, and in particular relates to a coking coal blending method based on intelligent optimization. Background Art

[0002] At present, the traditional manual coal blending method is used to specify the ratio of each raw coal and guide the coking production work, which can basically meet the daily production needs. However, due to the complex physical and chemical changes in the coking process and the complex influencing factors, the current coal blending coking results can only be roughly estimated based on manual experience, or qualitatively analyzed based on small coke oven tests, and it is difficult to accurately predict the product quality. Based on the coke quality requirements, it is difficult for manual coal blending to balance quality and price to obtain the optimal coal blending ratio, resulting in a more conservative value of the raw coal ratio, and the quality indicators are prone to excess, which increases the cost of coking. Under the actual production conditions of variable coal sources and fluctuating coal quality, it is difficult to quickly expand the scope of coal use and make timely adjustments to the coal blending structure based on manual experience alone. Therefore, it is necessary to optimize the coal blending process and use automated and intelligent methods instead of manual coal blending.

[0003] A blast furnace smelting cost management method with Chinese patent publication number CN110491454A, according to the parameters and proportions of coking coal, coking experiments are carried out respectively, coke parameters are obtained, and a mapping relationship between coke parameters and coking coal proportions is established. According to a series of relationships, the performance requirements of blast furnace smelting for coke are constrained, and a coking coal blending scheme that meets the cost requirements is calculated using a multivariate regression function. The above invention only collects samples through a large number of coal blending coking experiments to establish a database, and then uses regression and other methods to formulate a proportioning scheme, which only meets the set cost requirements or a simple single-objective proportioning optimization, and cannot reduce costs as much as possible to obtain the lowest cost scheme that meets the quality requirements. In addition, a large number of coal blending experiments and sample collection also generate high costs. In the face of the situation where coal sources are variable and coal quality fluctuations are more complex, the generalization ability is insufficient and better results cannot be obtained. Summary of the invention

[0004] (I) Purpose of the invention

[0005] In order to overcome the above shortcomings, the purpose of the present invention is to provide a coking coal blending method based on intelligent optimization to solve the above technical problems.

[0006] (II) Technical solution

[0007] To achieve the above objectives, the technical solutions provided by this application are as follows:

[0008] A method for coking coal blending based on intelligent optimization comprises the following steps:

[0009] S1 collects historical raw coal quality ratio data and corresponding coke quality data, performs data cleaning preprocessing and data standardization, and analyzes the correlation between various coke quality parameters and raw coal quality parameters;

[0010] S2 combines the correlation analysis results with historical data to optimize the hyperparameters of the base algorithm, determines the composition of each layer of the base algorithm of the multi-algorithm multi-level superposition model, and obtains the coke parameter prediction model set;

[0011] According to the coke quality constraints and on the basis of the coke parameter prediction model set, S3 establishes a multi-constraint single-objective stochastic optimization model to find the lowest cost matching scheme.

[0012] Preferably, when analyzing the corresponding relationship between raw coal quality and coke quality parameters in S1, the data to be collected include raw coal information, raw coal industrial analysis indicators, coal rock data and coke quality parameters.

[0013] The raw coal information includes name, type, price and proportion;

[0014] The industrial analysis indicators of the raw coal include moisture, ash, volatile matter, sulfur, bonding index, maximum colloid layer thickness, and final shrinkage;

[0015] Coke quality parameters include ash content, volatile matter, sulfur content, abrasion resistance, crushing strength, reactivity, and strength after thermal reaction.

[0016] Preferably, the preprocessing in S1 includes completing missing data, eliminating abnormal data, and processing duplicate data to ensure the integrity, validity, timeliness, consistency, accuracy and uniqueness of the data, and avoid the impact of data quality issues on subsequent processes.

[0017] Preferably, when analyzing the correlation between the coke quality parameter and the raw coal quality parameter in S1, the correlation between the coke quality parameter and the raw coal quality parameter is calculated by the Pearson correlation coefficient to obtain a raw coal-coke quality parameter correlation coefficient matrix:

[0018]

[0019] Among the coke quality parameters, Ad is ash content, Vdaf is volatile matter, Std is sulfur content, M10 is abrasion resistance, M25 and M40 represent crushing strength, CRI represents reactivity, and CSR is strength after thermal reaction;

[0020] Among the raw coal quality parameters, ad is ash content, vdaf is volatile matter, std is sulfur content, G is bonding index, Y is maximum gelatin layer thickness, and X is final shrinkage;

[0021] According to the correlation coefficient within the matrix, the raw coal quality parameters with strong correlation with each coke quality parameter are obtained.

[0022] Preferably, the multi-algorithm multi-level superposition model prediction in S2 specifically includes the following steps:

[0023] S21 obtains the raw coal quality parameters with the greatest correlation with each coke quality parameter according to the coke quality parameter correlation analysis result;

[0024] S22 builds a multi-algorithm and multi-level superposition model based on historical coal blending data for training and predicts various coke quality parameters;

[0025] S23 obtains the weighted value of each raw coal quality parameter according to the quality parameter and proportion of each raw coal, and then predicts Ad according to the weighted value coke value,

[0026]

[0027] Ad coke =F Ad (X)

[0028] In the formula, n is the code of raw coal type, rn is the ratio of raw coal n, ad coaln is the ash content of raw coal n, vdaf coaln is the volatile matter of raw coal n, std coaln is the sulfur content of raw coal n, G coaln is the bonding index of raw coal n, Y coaln is the maximum colloid layer thickness of raw coal n, X coaln is the final shrinkage of raw coal n, X is the weighted value of raw coal quality parameters, F(X) is a multi-algorithm multi-level superposition model, Ad coke Corresponding to the coke ash content.

[0029] Preferably, the F Ad (X) The multi-algorithm multi-level superposition model includes multiple base algorithms, including XGBoost, LightGBM, GBDT, GP, Adaboost, SVR, and RF, and specifically includes the following steps:

[0030] Step 1: train the base models respectively through historical coal blending data and apply Bayesian optimization of hyperparameters to obtain the trained base algorithms, and select the algorithms with good prediction effects to form the first layer of the multi-level superposition model;

[0031] Step 2, build the second-layer model through historical data, use the output of the first-layer algorithm as the input of the second-layer algorithm, conduct training and hyperparameter search, select the algorithm with better effect to form the second layer, and gradually build the third, fourth and nth layers. The number of layers of the model can be flexibly set, and the algorithm of each layer is set according to the prediction effect. Prediction is performed step by step through multiple layers and multiple algorithms, which greatly improves the prediction accuracy of coke quality parameters.

[0032] Preferably, the S3 specifically includes the following steps:

[0033] Randomly generate optimization factors, that is, various raw coal ratios:

[0034]

[0035] Establish the objective function:

[0036]

[0037] P=[p1 p2 p3 p4 ... p n ] T

[0038]

[0039] In the formula, U is the objective function value, that is, the cost value, is a single factor, P is the unit price of various types of raw coal;

[0040] Set up constraints:

[0041]

[0042]

[0043]

[0044] In the iterative process of optimization, the fitness 1 / U of each factor is calculated, and the existing factors are selected according to the size of the fitness. The larger the fitness, the greater the probability of being selected, so as to select the factors closer to the optimal value. In addition, the random generation probability pr is set, and a number between 0 and 1 is randomly generated in each iteration. If it is greater than pr, the current factor is retained. If it is less than pr, a new factor is randomly generated. Through multiple iterations, the factor that minimizes the objective function is extracted, which is the optimal ratio of raw coal, and the optimal ratio scheme with the lowest coal blending cost is obtained.

[0045] Preferably, the results of the coal blending optimization are displayed in the form of a table, in which the horizontal axis is the coal blending number, the vertical axis is the coal blending cost, and there are two lines, one for the coal blending cost and the other for the manual coal blending cost, which are represented by red and blue respectively.

[0046] Beneficial effects:

[0047] 1. The method of the present invention transforms manual coal blending into intelligent coal blending, trains the model according to historical data, and automatically generates the blending scheme through the model, which reduces manual participation, simplifies the process, and has lower blending cost and blending time.

[0048] 2. Analysis of the influence of coke quality parameters. The existing coking coal blending process does not analyze the influencing factors of a single coke quality parameter. This method uses Pearson correlation coefficient analysis to obtain the influence of raw coal quality parameters on each coke quality parameter. According to the analysis results, the coke quality parameters are predicted respectively to improve the prediction accuracy of coke quality.

[0049] 3. Multi-algorithm multi-level superposition model prediction. The model is composed of various machine learning and deep learning algorithms, and the optimal model parameters are obtained through hyper-parameter optimization. The model's flexible multi-level structure and multiple algorithm combinations reduce errors layer by layer and improve the prediction accuracy of coke quality parameters.

[0050] 4. Random optimization algorithm calculates the best ratio scheme. A random optimization algorithm is proposed. According to the raw coal quality parameters and coke quality constraints, multiple iterations are performed on the basis of the prediction model set to obtain the best ratio scheme, which is compared with the actual manual coal blending and has a lower cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 is a flow chart of the present invention;

[0052] Figure 2 It is a multi-algorithm multi-level superposition model diagram of the present invention;

[0053] Figure 3 It is a table comparing the coal blending cost of the method of the present invention with the cost of manual coal blending. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solutions and advantages of the present invention more clear, the following is a detailed description of the present invention in conjunction with the specific implementation methods and with reference to the attached drawings. Figure 1-3 , the present invention is further described in detail. It should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present invention. In addition, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessary confusion of the concept of the present invention.

[0055] The present invention provides a method for coking coal blending based on intelligent optimization, comprising the following steps:

[0056] S1 collects historical raw coal quality ratio data and corresponding coke quality data, performs data cleaning preprocessing and data standardization, and analyzes the correlation between various coke quality parameters and raw coal quality parameters;

[0057] The data that needs to be collected include raw coal information, raw coal industrial analysis indicators, coal rock data and coke quality parameters.

[0058] The raw coal information includes name, type, price and proportion;

[0059] The industrial analysis indicators of the raw coal include moisture, ash, volatile matter, sulfur, bonding index, maximum colloid layer thickness, and final shrinkage;

[0060] Coke quality parameters include ash content, volatile matter, sulfur content, abrasion resistance, crushing strength, reactivity, and strength after thermal reaction.

[0061] Preprocessing includes completing missing data, removing abnormal data, and processing duplicate data to ensure the integrity, validity, timeliness, consistency, accuracy, and uniqueness of the data, and to avoid the impact of data quality issues on subsequent operations.

[0062] When analyzing the correlation between the coke quality parameters and the raw coal quality parameters in S1, the correlation between the coke quality parameters and the raw coal quality parameters is calculated by the Pearson correlation coefficient to obtain the raw coal-coke quality parameter correlation coefficient matrix:

[0063]

[0064] Among the coke quality parameters, Ad is ash content, Vdaf is volatile matter, Std is sulfur content, M10 is abrasion resistance, M25 and M40 represent crushing strength, CRI represents reactivity, and CSR is strength after thermal reaction;

[0065] Among the raw coal quality parameters, ad is ash content, vdaf is volatile matter, std is sulfur content, G is bonding index, Y is maximum gelatin layer thickness, and X is final shrinkage;

[0066] According to the correlation coefficient within the matrix, the raw coal quality parameters with strong correlation with each coke quality parameter are obtained.

[0067] S2 combines the correlation analysis results with historical data to optimize the hyperparameters of the base algorithm, determines the composition of each layer of the base algorithm of the multi-algorithm multi-level superposition model, and obtains the coke parameter prediction model set;

[0068] S21 obtains the raw coal quality parameters with the greatest correlation with each coke quality parameter according to the coke quality parameter correlation analysis result;

[0069] S22 builds a multi-algorithm and multi-level superposition model based on historical coal blending data for training and predicts various coke quality parameters;

[0070] S23 obtains the weighted value of each raw coal quality parameter according to the quality parameter and proportion of each raw coal, and then predicts Ad according to the weighted value coke value,

[0071]

[0072] Ad coke =F Ad (X)

[0073] In the formula, n is the code of raw coal type, rn is the ratio of raw coal n, ad coaln is the ash content of raw coal n, vdaf coaln is the volatile matter of raw coal n, std coaln is the sulfur content of raw coal n, G coaln is the bonding index of raw coal n, Y coaln is the maximum colloid layer thickness of raw coal n, X coaln is the final shrinkage of raw coal n, X is the weighted value of raw coal quality parameters, F(X) is a multi-algorithm multi-level superposition model, Ad coke Corresponding to the coke ash content.

[0074] 8. The F Ad (X) The multi-algorithm multi-level superposition model includes multiple base algorithms, including XGBoost, LightGBM, GBDT, GP, Adaboost, SVR, and RF, and specifically includes the following steps:

[0075] Step 1: train the base models respectively through historical coal blending data and apply Bayesian optimization of hyperparameters to obtain the trained base algorithms, and select the algorithms with good prediction effects to form the first layer of the multi-level superposition model;

[0076] Step 2, build the second-layer model through historical data, use the output of the first-layer algorithm as the input of the second-layer algorithm, conduct training and hyperparameter search, select the algorithm with better effect to form the second layer, and gradually build the third, fourth and nth layers. The number of layers of the model can be flexibly set, and the algorithm of each layer is set according to the prediction effect. Prediction is performed step by step through multiple layers and multiple algorithms, which greatly improves the prediction accuracy of coke quality parameters.

[0077] According to the coke quality constraints and on the basis of the coke parameter prediction model set, S3 establishes a multi-constraint single-objective stochastic optimization model to find the lowest cost matching scheme.

[0078] The S3 specifically includes the following steps:

[0079] Randomly generate optimization factors, that is, various raw coal ratios:

[0080]

[0081] Establish the objective function:

[0082]

[0083] P=[p1 p2 p3 p4 ... p n ] T

[0084]

[0085] In the formula, U is the objective function value, that is, the cost value, is a single factor, P is the unit price of various types of raw coal;

[0086] Set up constraints:

[0087]

[0088]

[0089]

[0090] In the iterative process of optimization, the fitness 1 / U of each factor is calculated, and the existing factors are selected according to the size of the fitness. The larger the fitness, the greater the probability of being selected, so as to select the factors closer to the optimal value. In addition, the random generation probability pr is set, and a number between 0 and 1 is randomly generated in each iteration. If it is greater than pr, the current factor is retained. If it is less than pr, a new factor is randomly generated. Through multiple iterations, the factor that minimizes the objective function is extracted, which is the optimal ratio of raw coal, and the optimal ratio scheme with the lowest coal blending cost is obtained.

[0091] The results of the optimal coal ratio search are displayed in the form of a table, where the horizontal axis is the coal blending number and the vertical axis is the coal blending cost. There are two lines, one for the coal blending cost and the other for the manual coal blending cost, represented by red and blue respectively.

[0092] The present invention provides a coking coal blending method based on intelligent optimization. First, historical raw coal blending data and coke data are collected, data preprocessing is performed, and the correlation between coke quality and raw coal quality is analyzed; then, various machine learning and deep learning algorithms are used as base algorithms, and the hyperparameter optimization of the base algorithm is performed in combination with the correlation analysis results and historical data, and a multi-algorithm and multi-level superposition model of coke quality parameters is constructed to obtain a coke quality parameter prediction model set; finally, according to the coke quality constraints and the raw coal industrial analysis indicators, an optimization model based on a random optimization algorithm is established on the basis of the coke parameter prediction model set to obtain the lowest cost blending scheme.

[0093] The present invention has the following advantages:

[0094] The method of the present invention transforms manual coal blending into intelligent coal blending, trains a model based on historical data, and automatically generates a blending plan through the model, thereby reducing manual participation, simplifying the process, and having lower blending cost and blending time.

[0095] Analysis of the influence of coke quality parameters. The existing coking coal blending process does not analyze the influencing factors of a single coke quality parameter. This method uses Pearson correlation coefficient analysis to obtain the influence of raw coal quality parameters on each coke quality parameter. According to the analysis results, the coke quality parameters are predicted respectively to improve the prediction accuracy of coke quality.

[0096] Multi-algorithm multi-level superposition model prediction. The model is composed of various machine learning and deep learning algorithms, and the optimal model parameters are obtained through hyper-parameter optimization. The model's flexible multi-level structure and multiple algorithm combinations reduce errors layer by layer and improve the prediction accuracy of coke quality parameters.

[0097] The random optimization algorithm calculates the best ratio scheme. A random optimization algorithm is proposed. According to the raw coal quality parameters and coke quality constraints, multiple iterations are performed on the basis of the prediction model set to obtain the best ratio scheme, and compared with the actual manual coal blending, it has a lower cost.

[0098] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0099] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for coking coal blending based on intelligent optimization, characterized in that: The following steps are involved: S1 collects historical raw coal quality ratio data and corresponding coke quality data, performs data cleaning preprocessing and data standardization, and analyzes the correlation between various coke quality parameters and raw coal quality parameters; S2 combines the correlation analysis results with historical data to optimize the hyperparameters of the base algorithm, determines the composition of each layer of the base algorithm of the multi-algorithm multi-level superposition model, and obtains the coke parameter prediction model set; According to the coke quality constraints and on the basis of the coke parameter prediction model set, S3 establishes a multi-constraint single-objective stochastic optimization model to find the lowest cost matching scheme.

2. The method for coking coal blending based on intelligent optimization according to claim 1, characterized in that: In the above S1, when analyzing the corresponding relationship between raw coal quality and coke quality parameters, the data to be collected include raw coal information, raw coal industrial analysis indicators, coal rock data and coke quality parameters. The raw coal information includes name, type, price and proportion; The industrial analysis indicators of the raw coal include moisture, ash, volatile matter, sulfur, bonding index, maximum colloid layer thickness, and final shrinkage; Coke quality parameters include ash content, volatile matter, sulfur content, abrasion resistance, crushing strength, reactivity, and strength after thermal reaction.

3. The method for coking coal blending based on intelligent optimization according to claim 1, characterized in that: The preprocessing in S1 includes completing missing data, removing abnormal data, and processing duplicate data to ensure the integrity, validity, timeliness, consistency, accuracy and uniqueness of the data, and avoid the impact of data quality issues on subsequent operations.

4. The method for coking coal blending based on intelligent optimization according to claim 1, characterized in that: When analyzing the correlation between the coke quality parameters and the raw coal quality parameters in S1, the correlation between the coke quality parameters and the raw coal quality parameters is calculated by the Pearson correlation coefficient to obtain the raw coal-coke quality parameter correlation coefficient matrix: Among the coke quality parameters, Ad is ash content, Vdaf is volatile matter, Std is sulfur content, M10 is abrasion resistance, M25 and M40 represent crushing strength, CRI represents reactivity, and CSR is strength after thermal reaction; Among the raw coal quality parameters, ad is ash content, vdaf is volatile matter, std is sulfur content, G is bonding index, Y is maximum gelatin layer thickness, and X is final shrinkage; According to the correlation coefficient within the matrix, the raw coal quality parameters with strong correlation with each coke quality parameter are obtained.

5. The method for coking coal blending based on intelligent optimization according to claim 1, characterized in that: The multi-algorithm multi-level superposition model prediction in S2 specifically includes the following steps: S21 obtains the raw coal quality parameters with the greatest correlation with each coke quality parameter according to the coke quality parameter correlation analysis result; S22 builds a multi-algorithm and multi-level superposition model based on historical coal blending data for training and predicts various coke quality parameters; S23 obtains the weighted value of each raw coal quality parameter according to the quality parameter and proportion of each raw coal, and then predicts Ad according to the weighted value coke value, Ad coke =F Ad (X) In the formula, n is the code of raw coal type, rn is the ratio of raw coal n, ad coaln is the ash content of raw coal n, vdaf coaln is the volatile matter of raw coal n, std coaln is the sulfur content of raw coal n, G coaln is the bonding index of raw coal n, Y coaln is the maximum colloidal layer thickness of raw coal n, X coaln is the final shrinkage of raw coal n, X is the weighted value of raw coal quality parameters, F(X) is a multi-algorithm multi-level superposition model, Ad coke Corresponding to the coke ash content.

6. A method for coking coal blending based on intelligent optimization according to claim 5, characterized in that: The F Ad (X) The multi-algorithm multi-level superposition model includes multiple base algorithms, including XGBoost, LightGBM, GBDT, GP, Adaboost, SVR, and RF, and specifically includes the following steps: Step 1: train the base models respectively through historical coal blending data and apply Bayesian optimization of hyperparameters to obtain the trained base algorithms, and select the algorithms with good prediction effects to form the first layer of the multi-level superposition model; Step 2, build the second-layer model through historical data, use the output of the first-layer algorithm as the input of the second-layer algorithm, conduct training and hyperparameter search, select the algorithm with better effect to form the second layer, and gradually build the third, fourth and nth layers. The number of layers of the model can be flexibly set, and the algorithm of each layer is set according to the prediction effect. Prediction is performed step by step through multiple layers and multiple algorithms, which greatly improves the prediction accuracy of coke quality parameters.

7. A method for coking coal blending based on intelligent optimization according to claim 1, characterized in that: The S3 specifically includes the following steps: Randomly generate optimization factors, that is, various raw coal ratios: Establish the objective function: P=[p1 p2 p3 p4...p n ] T In the formula, U is the objective function value, that is, the cost value, is a single factor, P is the unit price of various types of raw coal; Set up constraints: In the iterative process of optimization, the fitness 1 / U of each factor is calculated, and the existing factors are selected according to the size of the fitness. The larger the fitness, the greater the probability of being selected, so as to select the factors closer to the optimal value. In addition, the random generation probability pr is set, and a number between 0 and 1 is randomly generated in each iteration. If it is greater than pr, the current factor is retained. If it is less than pr, a new factor is randomly generated. Through multiple iterations, the factor that minimizes the objective function is extracted, which is the optimal ratio of raw coal, and the optimal ratio scheme with the lowest coal blending cost is obtained.

8. A method for coking coal blending based on intelligent optimization according to claim 7, characterized in that: The results of the optimal coal ratio search are displayed in the form of a table, where the horizontal axis is the coal blending number and the vertical axis is the coal blending cost. There are two lines, one for the coal blending cost and the other for the manual coal blending cost, represented by red and blue respectively.

Images