Precision coking coal blending method and system

Abstract

The present application relates to the field of coal coking technology, and in particular to a precise coking coal blending method and system, which obtains the index of a single coking coal to establish a single coking coal resource information database; based on the single coking coal resource information database, a blending coal quality prediction model is established according to a linear model and an SVR fusion model to obtain the blending coal quality index; based on the blending coal quality index and a historical coking process data set, a coke quality prediction model is established through a KAN neural network to obtain the coke quality index; based on the single coal quality parameters and coking process parameters, a sampling ratio optimization model is established with expected blending coal quality index, expected coke quality index and inventory as constraints to obtain candidate ratio schemes; candidate ratio schemes are screened based on the blending coal quality prediction model, the coke quality prediction model and the blending coal cost to obtain the optimal coal blending scheme, thereby solving the problem of inaccurate coking coal blending methods.

Description

Technical Field

[0001] The present application relates to the field of coal coking technology, and in particular to a method and system for precise coking coal blending. Background Art

[0002] Coking coal is a key raw material for coke production in steel companies. Stable coke quality is crucial for blast furnace ironmaking. Coke quality is determined by the coking process, and coking coal blending is a key step in the coking process. This involves mixing different types and proportions of coal to produce coke. Commonly used coal types include fat coal, coking coal, lean coal, and 1 / 3 coking coal. Properly blending these coal types optimizes the physical and chemical properties of coke to meet the needs of various industrial processes.

[0003] The quality of coal blending technology directly impacts coke quality and cost, making it a critical issue that coking enterprises urgently need to address. Traditional coal blending methods rely on the experience of blenders and lack systematic scientific guidance. This leads to low production efficiency, long calculation cycles, and unstable coke quality. The difficulty in translating experience into data also prevents employees from mastering coal blending technology quickly. The complexity and variability of production sites make it difficult for empirical coal blending to quickly adapt to changing demand, resulting in frequent adjustments and increased costs.

[0004] With technological advancements and increasing market demand, companies are placing higher demands on coke quality and stability. Coal blending methods in related technologies primarily focus on coke quality prediction. However, even when only coke quality prediction is met, coal blending still suffers from inaccuracy. Summary of the Invention

[0005] The present application provides a precise coking coal blending method and system to solve the problem of inaccurate coking coal blending methods.

[0006] In a first aspect, the present application provides a method for precise coking coal blending, the method comprising:

[0007] Obtain indicators of a single type of coking coal to establish a single type of coking coal resource information database;

[0008] Based on the single coking coal resource information database, a blending coal quality prediction model is established according to a linear model and an SVR fusion model to obtain blending coal quality indicators;

[0009] Based on the blended coal quality index and the historical coking process data set, a coke quality prediction model is established through a KAN neural network to obtain the coke quality index;

[0010] Based on the quality parameters of individual coals and coking process parameters, and with the expected quality index of blended coal, expected quality index of coke and inventory as constraints, a sampling ratio optimization model is established to obtain candidate ratio schemes;

[0011] The candidate blending schemes are screened based on the blending coal quality prediction model, the coke quality prediction model and the blending coal cost to obtain the optimal coal blending scheme.

[0012] The above method first analyzes production data to identify the correlation between various coke quality indicators and the quality indicators of individual coal types, thereby establishing a coke quality prediction model. Then, based on the coke quality requirements, the blending coal quality requirements, and constraints such as the quality, price, and inventory of each individual coal type, the optimal ratio of the required individual coal types and each type of coal is determined to achieve the optimal coal blending solution. This method not only intelligently predicts various coke quality indicators based on the characteristics and ratios of each individual coal type, but also intelligently and quickly calculates the optimal coal blending solution based on actual production needs and the characteristics of each individual coal type. This improves coke quality stability and production efficiency, reduces production costs, and addresses the issue of inaccurate coking coal blending methods.

[0013] Optionally, the indicators of a single type of coking coal include: maximum thickness of the gelatinous layer, bonding index, coal rock indicators of a single type of coal, and work point data.

[0014] Optionally, the step of establishing a blended coal quality prediction model based on the single coking coal resource information database and according to a linear model and an SVR fusion model includes:

[0015] Generate a single coal derived feature MBI through the single coking coal resource information database;

[0016] Based on the single coking coal resource information database, single coal derived characteristics MBI and coal blending list, and according to the linear model and SVR fusion model, a blending coal quality prediction model is established.

[0017] The above method generates a single-type coal derived feature MBI from the single-type coking coal resource information database, which can deeply explore the coal quality characteristics of a single type of coal and extract a comprehensive index that better reflects the key characteristics of a single type of coal in the coking process. The use of a linear model and an SVR fusion model to construct a coordinated coal quality prediction model can comprehensively leverage the advantages of both models. It can not only quickly capture linear relationships in the data, but also use kernel functions to map nonlinear problems in low-dimensional input space to high-dimensional feature space, so that linear models can be used to fit data in high-dimensional feature space, effectively processing data that originally exhibited nonlinear relationships in low-dimensional space, thereby effectively fitting these nonlinear relationships and improving the model's prediction accuracy.

[0018] Optionally, the step of establishing a coke quality prediction model based on the blended coal quality index and the historical coking process data set by using a KAN neural network includes:

[0019] Screening out historical coking process data samples from the historical coking process data set; the historical coking process data samples include coking process condition parameters corresponding to the blended coal produced each time and historical coke quality indicators;

[0020] A robust normalization method is used to scale the blended coal quality index and coking process condition parameters to obtain a model feature sample set;

[0021] Establishing a coke quality prediction data set based on the model feature sample set and the historical coke quality indicators;

[0022] A coke quality prediction model is established using the coke quality prediction data set and the KAN neural network.

[0023] The above method can reduce the impact of outliers through sample screening and robust normalization, effectively handle variable dimensional differences, balance variable contributions, reduce the impact of noise, and improve model stability. The coke quality prediction model established by combining the coke quality prediction dataset and the KAN neural network can automatically learn inherent patterns from the data and adapt to different datasets and problems by adjusting the network structure and parameters. It has good flexibility and adaptability to achieve accurate prediction of coke quality.

[0024] Optionally, the input variables of the coke quality prediction model include adhesion index, volatile matter, moisture, ash content, sulfur content, coking time, coke side temperature and ash composition; the output of the coke quality prediction model includes the ash content, sulfur content, crushing strength, abrasion resistance, reactivity index and post-reaction strength of the coke.

[0025] The input variables and outputs of the coke quality prediction model can accurately consider coal quality, coking process and ash composition, comprehensively evaluate coke quality, meet the needs of different application scenarios, and provide strong support for coking process optimization and quality control.

[0026] Optionally, the steps for obtaining the optimal coal blending solution include:

[0027] Taking the inventory of each type of coal and the sum of all proportions as one as constraints, a sampling proportion optimization model is established to obtain candidate proportion schemes;

[0028] Inputting the candidate blending scheme, single coal quality parameters, and coking process parameters into the blended coal quality prediction model and the coke quality prediction model to screen out feasible solutions that meet the blended coal quality constraints and the coke quality constraints;

[0029] The price of the feasible solution is calculated, and the optimal coal blending solution is selected with the lowest blending coal cost as the objective function.

[0030] This method not only completely resolves the problem of non-uniform blending ratios, but also intelligently and rapidly calculates the optimal blending ratio that satisfies the constraints. It can more quickly and accurately find a solution where the sum of the blending ratios is 1 and the inventory limit is met, especially for situations where the inventory ratio slightly exceeds the coal blending requirements. It also accommodates the selection of various blending ratio calculation accuracies, enabling comparison of changes in coal quality, coke quality, and cost at different calculation accuracies while meeting the requirements of coal blenders.

[0031] Optionally, the objective function is calculated as follows:

[0032] ;

[0033] Among them, n is the total amount of single type of coal used in coal blending; For the The cost price of a single type of coal; For the The ratio of a single type of coal.

[0034] Optionally, the calculation formula of the sampling ratio optimization model includes:

[0035] ;

[0036] Where n is the total amount of single type of coal used in coal blending; C is the cost price of each single type of coal; The proportion of each single type of coal; W To match coal quality indicators; V It is the quality index of coke; For the The proportion of a single type of coal; h j For the The inventory ratio of a single type of coal.

[0037] Optionally, the quality constraints of the blended coal include the bonding index, gelatinous layer thickness and volatile matter of the blended coal; the quality constraints of the coke include ash content, sulfur content, crushing strength, wear resistance, reactivity index and post-reaction strength.

[0038] By strictly controlling the bonding index, gelatinous layer thickness and volatile matter of the blended coal, it is possible to ensure that the coal can be fully softened, melted and solidified during the coking process, thereby improving the cold strength, thermal stability and overall quality of the coke. By limiting the ash and sulfur content of the coke, the impurity content can be reduced and the harmful components in the coke can be reduced. The blended coal quality constraints and the coke quality constraints can also help guide the design of the coal blending ratio, making the selection of raw coal more scientific and reasonable, thereby optimizing the coking process parameters and improving production efficiency. By controlling indicators such as crushing strength, wear resistance, reactivity index and post-reaction strength, it is beneficial to ensure the behavior of coke during the blast furnace smelting process, reduce powder generation, improve the permeability of the charge, and improve smelting efficiency. By controlling the volatile matter and sulfur content, it is helpful to reduce the emission of pollutants in the coking process, especially the emission of sulfides and nitrogen oxides, to meet environmental protection requirements and reduce the environmental liability risk of the enterprise.

[0039] A second aspect of the present application provides a precise coking coal blending system, applicable to the precise coking coal blending method described in the first aspect, the system comprising:

[0040] A resource information database construction module is used to obtain the indicators of a single type of coking coal to establish a single type of coking coal resource information database;

[0041] A blending coal quality prediction model building module is used to establish a blending coal quality prediction model based on the single coking coal resource information database and according to a linear model and an SVR fusion model to obtain blending coal quality indicators;

[0042] A coke quality prediction model building module is used to establish a coke quality prediction model based on the blend coal quality index and the historical coking process data set and through a KAN neural network to obtain the coke quality index;

[0043] The proportion optimization module is used to establish a sampling proportion optimization model based on the quality parameters of a single coal and the coking process parameters, and with the expected blending coal quality indicators, expected coke quality indicators and inventory as constraints to obtain candidate proportion schemes; and screen the candidate proportion schemes based on the blending coal quality prediction model, the coke quality prediction model and the blending coal cost to obtain the optimal coal blending scheme.

[0044] The system first analyzes production data to identify correlations between various coke quality indicators and those of individual coal types, thereby establishing a coke quality prediction model. It then determines the optimal ratio of the required individual coal types and each type of coal, based on coke quality requirements, blending coal quality requirements, and constraints such as the quality, price, and inventory of each individual coal type, to arrive at the optimal coal blending solution. This method not only intelligently predicts various coke quality indicators based on the characteristics and proportions of each individual coal type, but also intelligently and rapidly calculates the optimal coal blending solution based on actual production needs and the characteristics of each individual coal type. This improves coke quality stability and production efficiency, reduces production costs, and addresses the issue of inaccurate coking coal blending methods.

[0045] It can be seen from the above technical solutions that the present application provides a precise coking coal blending method and system, which establishes a single coking coal resource information database by acquiring the indicators of a single coking coal; based on the single coking coal resource information database, a blending coal quality prediction model is established according to the linear model and the SVR fusion model to obtain the blending coal quality indicators; based on the blending coal quality indicators and the historical coking process data set, a coke quality prediction model is established through the KAN neural network to obtain the coke quality indicators; based on the single coal quality parameters and the coking process parameters, a sampling ratio optimization model is established with the expected blending coal quality indicators and the expected coke quality indicators and the inventory as constraints to obtain candidate ratio schemes; based on the blending coal quality prediction model, the coke quality prediction model and the blending coal cost, the candidate ratio schemes are screened to obtain the optimal coal blending scheme to solve the problem of inaccurate coking coal blending methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solution of the present application, the following is a brief introduction to the drawings required for use in the embodiments. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0047] Figure 1 This is a flow chart of the precise coking coal blending method described in the embodiments of the present application;

[0048] Figure 2 Schematic diagram of the process of establishing a blended coal quality prediction model in the precise coking coal blending method described in the embodiment of the present application;

[0049] Figure 3 Schematic diagram of the process of establishing a coke quality prediction model in the precise coking coal blending method described in the embodiment of the present application;

[0050] Figure 4 Schematic diagram of the calculation process of the sampling ratio optimization model in the precise coking coal blending method described in the embodiment of this application. DETAILED DESCRIPTION

[0051] The following embodiments are described in detail, with examples illustrated in the accompanying drawings. When the following description refers to the drawings, identical numerals in different figures represent identical or similar elements unless otherwise indicated. The embodiments described in the following embodiments are not intended to represent all possible implementations consistent with the present application. They are merely examples of systems and methods consistent with certain aspects of the present application.

[0052] To solve the problem of inaccurate coking coal blending method, see Figure 1 Some embodiments of this application provide a method for accurate coking coal blending, the method comprising:

[0053] S100: Obtain indicators of a single type of coking coal to establish a single type of coking coal resource information database.

[0054] It should be understood that by entering the single coking coal index into the resource information database through multi-system integration, a single coking coal resource information database can be established. This single coking coal resource information database enables centralized information management, orderly storage, and convenient query, access, and update of the single coking coal index, providing a solid data foundation for subsequent analysis and model building.

[0055] In some embodiments, the indicators of the single type of coking coal include: maximum thickness of the gelatinous layer, bonding index, coal rock indicators of the single type of coal, and work point data.

[0056] It should be understood that when bituminous coal is heated, a layer of viscous liquid and semi-liquid material, called colloid, forms on its surface. As the temperature rises further, the colloid solidifies and forms a solid substance, namely coke. The maximum thickness of the colloid layer (Y value) refers to the maximum thickness of the colloid formed by bituminous coal under specific conditions. The maximum thickness of the colloid layer varies significantly among different coal types. For example, fat coal has a large number of colloids and high fluidity, and its Y value is generally higher, typically between 20mm and 40mm. In contrast, gas coal and coking coal have relatively low Y values, with Y values ​​of less than 15mm for gas coal and between 10mm and 25mm for coking coal. A thicker colloid layer generally indicates that the coal has good cohesive properties, which contribute to the formation of structurally stable coke.

[0057] The cohesiveness index (G value) is an important indicator of coal's cohesiveness. It reflects the coal's ability to form coke during heating. Specifically, a certain amount of coal sample is mixed evenly with specialized anthracite, heated in a specific experimental device, and the strength of the resulting coke is used to calculate the G value. Coal with a high G value is better able to bond together, forming a coke mass. This improves the mechanical strength and stability of the coke, facilitating subsequent processes such as blast furnace smelting.

[0058] The petrographic index (Rmax) of a single coal type refers to the maximum expansion pressure it can reach during heating. It is a key indicator of coal's expansion and plasticity. By measuring the expansion pressure of coal at different temperatures, an expansion pressure curve can be developed, which in turn determines the Rmax value. Understanding the Rmax values ​​of different coal types allows accurate understanding of the differences in metamorphic degree among the various coals, allowing for optimal blending and adjustment of the various coal types. For example, when producing coke of a specific strength and quality, the Rmax values ​​of the individual coal types can be used to select the appropriate blend of coking coal and thermal coal to ensure consistent coke quality. Rmax can also, to a certain extent, predict coke quality indicators such as strength and reactivity. Generally speaking, within a certain Rmax range, coke has relatively low reactivity and high mechanical strength. The Rmax values ​​of individual coal types can be used to develop mathematical models or empirical formulas to predict the quality of coke after blending.

[0059] Coal analysis data refers to the industrial analysis of coal, including the contents of components such as moisture, ash, volatile matter, and fixed carbon. These data are crucial for assessing coal quality and suitability. For example, low ash content and a suitable volatile matter content are typically characteristics of high-quality coking coal.

[0060] S200: Based on the single coking coal resource information database, a blending coal quality prediction model is established according to a linear model and an SVR fusion model to obtain a blending coal quality index.

[0061] In some embodiments, see Figure 2 The steps of establishing a blended coal quality prediction model based on the single coking coal resource information database and the linear model and the SVR fusion model include:

[0062] S210: Generate a single-type coal derived feature MBI through the single-type coking coal resource information database.

[0063] It should be understood that the Mineral Base Index (MBI), a characteristic derived from a single coal type, characterizes the alkaline metal content in the coal ash. Higher MBI values ​​generally indicate a higher coke reactivity (CRI) because alkaline metals in the coal ash catalyze the reaction between coke and carbon dioxide. As the MBI value increases, the coke's structure is further damaged by carbon dioxide and alkali metal attack, resulting in a decrease in its post-reaction strength (CSR). Higher MBI values ​​further damage the coke's internal structure during the coking process, reducing its wear resistance, manifested as increased wear resistance. Higher MBI values ​​generally reduce the coke's crushing strength because alkaline metals penetrate the coke's carbon structure, causing deformation and cracking. Therefore, coke with a lower MBI value is less reactive. When used in blast furnaces, it can better withstand high temperatures and chemical attack, maintaining high strength and stability, thereby facilitating smooth blast furnace operation and extending the coke's service life. Coking companies can develop more scientific and reasonable coal blending plans based on the MBI values ​​of different individual coals, combined with other coal quality indicators. By adjusting the coal blending ratio, the MBI value of the blended coal can be reduced, improving the quality of the coke while also rationally utilizing resources and reducing production costs.

[0064] The calculation formula of the single coal derived characteristic MBI can be:

[0065] .

[0066] in, ω(A d ） is the ash content in a single type of coal; ω(K 2 O） is the potassium oxide content in a single type of coal; ω(Na 2 O） is the content of sodium oxide in a single type of coal; ω (CaO) is the calcium oxide content in a single type of coal; ω (MgO) is the content of magnesium oxide in a single type of coal; ω(Fe 2 O 3 ） is the content of ferric oxide in a single type of coal; ω(V daf ） is the volatile matter content in a single type of coal; ω (SiO 2 ） is the content of silicon dioxide in a single type of coal; ω(Al 2 O 3） It is the content of aluminum oxide in a single type of coal.

[0067] S220: Based on the single coking coal resource information database, the single coal derived characteristics MBI and the coal blending list, and according to the linear model and the SVR fusion model, a blended coal quality prediction model is established.

[0068] It should be understood that in the process of predicting coke quality, it is necessary to predict the quality indicators of the blended coal by using the quality indicators and proportions of individual coals, and thus predict the quality indicators of the coke. Different coal types play different roles in the blended coal. The coking properties of the blended coal depend on the performance and proportions of each individual coal type. The coking properties of the blended coal are the result of the combined effects of the coking properties of each individual coal type. By fully leveraging the characteristics of each coal type and complementing each other's strengths, the quality of coke can be improved. While linear models are used to predict the quality of blended coal, the quality of individual coal types within a batch is prone to fluctuations, and the linear model needs to be modified based on the test results of the blended coal. Therefore, it is necessary to integrate linear models with SVR (Support Vector Regression) models to predict the quality of blended coal and improve the accuracy of blended coal quality prediction.

[0069] The calculation formula of the linear model is:

[0070] ;

[0071] in, is the dependent variable vector, i.e., the observed value; is the observation matrix of independent variables, each column is an independent variable; is a vector containing the parameters that are usually to be estimated; is the error term, which represents the influence of factors other than the independent variable on the dependent variable.

[0072] The calculation formula of the SVR fusion model is:

[0073] ;

[0074] in, and is the Lagrange multiplier; is the sum function, and b is the bias term.

[0075] The above method generates a single-type coal derived feature MBI from the single-type coking coal resource information database, which can deeply explore the coal quality characteristics of a single type of coal and extract a comprehensive index that better reflects the key characteristics of a single type of coal in the coking process. The use of a linear model and an SVR fusion model to construct a coordinated coal quality prediction model can comprehensively leverage the advantages of both models. It can not only quickly capture linear relationships in the data, but also use kernel functions to map nonlinear problems in low-dimensional input space to high-dimensional feature space, so that linear models can be used to fit data in high-dimensional feature space, effectively processing data that originally exhibited nonlinear relationships in low-dimensional space, thereby effectively fitting these nonlinear relationships and improving the model's prediction accuracy.

[0076] S300: Based on the blended coal quality index and the historical coking process data set, a coke quality prediction model is established through a KAN neural network to obtain a coke quality index.

[0077] It should be understood that the KAN neural network is an innovative neural network architecture. Unlike traditional neural networks that use fixed activation functions, it can use learnable activation functions at the edge of the network. This design allows each weight parameter in the KAN to be replaced by a single variable function. These functions are usually parameterized in the form of spline functions, which provides extremely high flexibility and can simulate complex functions with fewer parameters. This is not only efficient and accurate, but also enhances the interpretability of the model. Therefore, when using the KAN neural network to predict the quality indicators of coke, the calculation formula of the KAN neural network can be expressed as follows:

[0078] ;

[0079] in, and is a single variable continuous function, and The calculation formulas are:

[0080] ;

[0081] .

[0082] Among them, the B-spline mechanism is the most important core learning mechanism of the KAN neural network. They replace the traditional weight parameters often used in neural networks. Their flexibility enables them to adaptively fit complex relationships in the data by adjusting their shape, thereby minimizing approximation errors and enhancing the network's ability to learn subtle patterns from high-dimensional data sets. B-splines can be expressed as:

[0083] ;

[0084] in, are the coefficients optimized during training, It is a B-spline basis function defined on the grid and is a hyperparameter that affects the accuracy of the network.

[0085] In some embodiments, the input variables of the coke quality prediction model include adhesiveness index, volatile matter, moisture, ash, sulfur, coking time, coke side temperature and ash composition; the output of the coke quality prediction model includes the ash content, sulfur content, crushing strength, abrasion resistance, reactivity index and post-reaction strength of the coke.

[0086] It should be understood that the input cohesiveness directly affects whether the coal can form coke of sufficient strength during the coking process. Indicators such as volatile matter, moisture, ash, and sulfur reflect the composition of the coal and have a significant impact on coke yield, quality, and furnace life. Coking time determines the residence time of the coal in the coke oven, directly affecting the maturity and structure of the coke. The coke side temperature influences the pyrolysis and solidification processes of the coal and plays a key role in the properties of the coke. Considering ash composition allows for more precise predictions of coke quality characteristics and provides more accurate results. Output ash and sulfur content are important indicators of coke purity and directly affect its performance in blast furnace smelting. Crushing strength and abrasion resistance reflect the coke's resistance to breakage and wear during handling and use. Reactivity index and post-reaction strength are closely related to the behavior of coke in chemical reactions and have a significant impact on the reaction processes and efficiency within the blast furnace, ensuring normal operation and efficient production.

[0087] The input variables and outputs of the coke quality prediction model can accurately consider coal quality, coking process and ash composition, comprehensively evaluate coke quality, meet the needs of different application scenarios, and provide strong support for coking process optimization and quality control.

[0088] In some embodiments, see Figure 3 The steps of establishing a coke quality prediction model based on the blended coal quality index and the historical coking process data set by using a KAN neural network include:

[0089] S310: Filter out historical coking process data samples from the historical coking process data set.

[0090] The historical coking process data samples include the coking process parameters and historical coke quality indicators corresponding to each batch of coal produced. Due to the presence of various interference factors in industrial sites, it is necessary to remove some samples that are obviously inconsistent with physical meaning from the historical coking process data set to ensure that the historical coking process data samples do not contain abnormal data.

[0091] S320: Using a robust normalization method to perform variable scaling on the blended coal quality index and coking process condition parameters to obtain a model feature sample set.

[0092] It should be understood that the robust normalization method is a normalization technique used to deal with outliers in data preprocessing, which aims to make the data insensitive to outliers. Due to the different dimensions of the blended coal quality indicators and coking process condition parameters, the sizes of different variable data vary greatly. At the same time, the different data distribution ranges will exaggerate the impact of certain variables on the target and mask the contribution of certain variables. Due to the presence of some noise in the data, directly applying min-max normalization and standardization methods will also affect the accuracy of the model. Therefore, the robust normalization method can be used to scale the variables of the blended coal quality indicators and coking process condition parameters, which can effectively reduce the impact of outliers on the normalization results and better maintain the distribution characteristics of the data.

[0093] S330: Establishing a coke quality prediction data set based on the model feature sample set and the historical coke quality indicators.

[0094] S340: Establishing a coke quality prediction model using the coke quality prediction dataset and the KAN neural network.

[0095] It should be understood that the coke quality prediction is performed using a KAN neural network, with coal quality indicators and coking process condition parameters as input and coke quality indicators as output.

[0096] The above method can reduce the impact of outliers through sample screening and robust normalization, effectively handle variable dimensional differences, balance variable contributions, reduce the impact of noise, and improve model stability. The coke quality prediction model established by combining the coke quality prediction dataset and the KAN neural network can automatically learn inherent patterns from the data and adapt to different datasets and problems by adjusting the network structure and parameters. It has good flexibility and adaptability to achieve accurate prediction of coke quality.

[0097] S400: Based on the quality parameters of a single type of coal and the coking process parameters, and with the expected quality index of the blended coal, the expected quality index of the coke and the inventory as constraints, a sampling ratio optimization model is established to obtain candidate ratio schemes.

[0098] It should be understood that the coking process parameters include coking time and coke side temperature.

[0099] S500: Screening the candidate blending schemes based on the blending coal quality prediction model, the coke quality prediction model and the blending coal cost to obtain an optimal coal blending scheme.

[0100] The above method first analyzes production data to identify the correlation between various coke quality indicators and the quality indicators of individual coal types, thereby establishing a coke quality prediction model. Then, based on the coke quality requirements, the blending coal quality requirements, and constraints such as the quality, price, and inventory of each individual coal type, the optimal ratio of the required individual coal types and each type of coal is determined to achieve the optimal coal blending solution. This method not only intelligently predicts various coke quality indicators based on the characteristics and ratios of each individual coal type, but also intelligently and quickly calculates the optimal coal blending solution based on actual production needs and the characteristics of each individual coal type. This improves coke quality stability and production efficiency, reduces production costs, and addresses the issue of inaccurate coking coal blending methods.

[0101] In some embodiments, the step of obtaining the optimal coal blending solution includes:

[0102] The inventory of each type of coal and the sum of all proportions are taken as constraints to establish a sampling proportion optimization model to obtain candidate proportion schemes.

[0103] The candidate blending schemes, single coal quality parameters and coking process parameters are input into the blended coal quality prediction model and the coke quality prediction model to screen out feasible solutions that meet the blended coal quality constraints and the coke quality constraints.

[0104] The price of the feasible solution is calculated, and the optimal coal blending solution is selected with the lowest blending coal cost as the objective function.

[0105] This method not only completely resolves the problem of non-uniform blending ratios, but also intelligently and rapidly calculates the optimal blending ratio that satisfies the constraints. It can more quickly and accurately find a solution where the sum of the blending ratios is 1 and the inventory limit is met, especially for situations where the inventory ratio slightly exceeds the coal blending requirements. It also accommodates the selection of various blending ratio calculation accuracies, enabling comparison of changes in coal quality, coke quality, and cost at different calculation accuracies while meeting the requirements of coal blenders.

[0106] In some embodiments, the objective function is calculated as follows:

[0107] ;

[0108] Among them, n is the total amount of single type of coal used in coal blending; For the The cost price of a single type of coal; For the The ratio of a single type of coal.

[0109] In some embodiments, the calculation formula of the sampling ratio optimization model includes:

[0110] ;

[0111] Where n is the total amount of single type of coal used in coal blending; C is the cost price of each single type of coal; is the proportion of each single coal; W is the quality index of the blended coal; V is the quality index of the coke; For the The ratio of a single coal; hj is the The inventory ratio of a single type of coal.

[0112] It should be understood that V min The minimum value of coke quality index ，V max The maximum value of coke quality index ，W max To match the maximum value of coal quality index ，W min To match the minimum values ​​of coal quality indicators.

[0113] See also Figure 4 The calculation process of the sampling ratio optimization model mainly includes the following 7 steps:

[0114] (1) Initialization parameters: unfinished proportion xn = 1; completed proportion xy = 0; inventory of a single type of coal KC = [kc1, kc2, …, kcn]; remaining inventory xs = sum(KC).

[0115] (2) Update the solution. Randomly and uniformly generate the quantity x = xi of the i-th type of coal from the inventory KC of a single type of coal. Calculate the unfinished proportion xn = xn-xi, the completed proportion xy = xy+xi, and the remaining inventory xs = xs-kci.

[0116] (3) Judgment condition xn-xs>0 or xy>1;

[0117] (4) If condition (3) is not satisfied, then i=i+1, and determine whether i is equal to the maximum number of iterations m. If so, output the candidate solution (i.e., the candidate matching scheme);

[0118] (5) If condition (3) is met, calculate the unfinished ratio xn = xn-xi, the completed ratio xy = xy-xi, and the remaining inventory xs = xs+kci.

[0119] (6) Update the solution according to the method of updating the solution in (2);

[0120] (7) Execute step (3) and repeat the calculation until a candidate solution (i.e., a candidate matching scheme) is output.

[0121] In some embodiments, the quality constraints of the blended coal include the bonding index, gelatinous layer thickness and volatile matter of the blended coal; the quality constraints of the coke include ash content, sulfur content, crushing strength, wear resistance, reactivity index and post-reaction strength.

[0122] Strictly controlling the cohesiveness index, gelatinous layer thickness, and volatile matter content of the blended coal ensures sufficient softening, melting, and solidification of the coal during the coking process, thereby improving the cold strength, thermal stability, and overall quality of the coke. Limiting the ash and sulfur content of the coke reduces impurity levels and harmful components in the coke. These blended coal and coke quality constraints also help guide the design of the coal blending ratio, enabling more rational selection of raw coals and optimizing coking process parameters (such as coke-side temperature and coking time), thereby improving production efficiency. Controlling parameters such as crushing strength, abrasion resistance, reactivity index, and post-reaction strength helps ensure coke behavior during blast furnace smelting, reduces fines generation, improves charge permeability, and enhances smelting efficiency. Controlling volatile matter and sulfur content helps reduce pollutant emissions during the coking process, particularly sulfides and nitrogen oxides, to comply with environmental protection requirements and mitigate the company's environmental liability risks.

[0123] Some embodiments of the present application further provide a precise coking coal blending system, which is applicable to the precise coking coal blending method described in the above embodiments, and the system comprises:

[0124] The resource information database construction module is used to obtain the indicators of a single type of coking coal to establish a single type of coking coal resource information database.

[0125] The blending coal quality prediction model construction module is used to establish a blending coal quality prediction model based on the single coking coal resource information database and according to the linear model and the SVR fusion model to obtain the blending coal quality index.

[0126] The coke quality prediction model building module is used to establish a coke quality prediction model based on the blended coal quality index and the historical coking process data set and through the KAN neural network to obtain the coke quality index.

[0127] The proportion optimization module is used to establish a sampling proportion optimization model based on the quality parameters of a single coal and the coking process parameters, and with the expected blending coal quality indicators, expected coke quality indicators and inventory as constraints to obtain candidate proportion schemes; and screen the candidate proportion schemes based on the blending coal quality prediction model, the coke quality prediction model and the blending coal cost to obtain the optimal coal blending scheme.

[0128] The system first analyzes production data to identify correlations between various coke quality indicators and those of individual coal types, thereby establishing a coke quality prediction model. It then determines the optimal ratio of the required individual coal types and each type of coal, based on coke quality requirements, blending coal quality requirements, and constraints such as the quality, price, and inventory of each individual coal type, to arrive at the optimal coal blending solution. This method not only intelligently predicts various coke quality indicators based on the characteristics and proportions of each individual coal type, but also intelligently and rapidly calculates the optimal coal blending solution based on actual production needs and the characteristics of each individual coal type. This improves coke quality stability and production efficiency, reduces production costs, and addresses the issue of inaccurate coking coal blending methods.

[0129] It can be seen from the above technical solutions that the embodiments of the present application provide a precise coking coal blending method and system, which establishes a single coking coal resource information database by acquiring the indicators of a single coking coal; based on the single coking coal resource information database, a blending coal quality prediction model is established according to the linear model and the SVR fusion model to obtain the blending coal quality indicators; based on the blending coal quality indicators and the historical coking process data set, a coke quality prediction model is established through the KAN neural network to obtain the coke quality indicators; based on the single coal quality parameters and the coking process parameters, a sampling proportion optimization model is established with the expected blending coal quality indicators and the expected coke quality indicators and the inventory as constraints to obtain candidate proportioning schemes; based on the blending coal quality prediction model, the coke quality prediction model and the blending coal cost, the candidate proportioning schemes are screened to obtain the optimal coal blending scheme to solve the problem of inaccurate coking coal blending methods.

[0130] Similar parts between the embodiments provided in this application can be referenced to each other. The specific implementation methods provided above are only a few examples under the overall concept of this application and do not constitute a limitation on the scope of protection of this application. For those skilled in the art, any other implementation methods expanded based on the scheme of this application without expending creative work shall fall within the scope of protection of this application.

Claims

1. A method for precise coking coal blending, characterized in that: The method comprises: Obtain indicators of a single type of coking coal to establish a single type of coking coal resource information database; Based on the single coking coal resource information database, a blending coal quality prediction model is established according to a linear model and an SVR fusion model to obtain blending coal quality indicators; Based on the blended coal quality index and the historical coking process data set, a coke quality prediction model is established through a KAN neural network to obtain the coke quality index; Based on the quality parameters of individual coals and coking process parameters, and with the expected quality index of blended coal, expected quality index of coke and inventory as constraints, a sampling ratio optimization model is established to obtain candidate ratio schemes; Screening the candidate blending schemes based on the blending coal quality prediction model, the coke quality prediction model, and the blending coal cost to obtain an optimal coal blending scheme; The steps of establishing a blended coal quality prediction model based on the single coking coal resource information database and according to the linear model and the SVR fusion model include: Generate a single coal derived feature MBI through the single coking coal resource information database; Based on the single coking coal resource information database, single coal derived characteristics MBI and coal blending list, a blending coal quality prediction model is established according to the linear model and SVR fusion model; The steps of establishing a coke quality prediction model based on the blended coal quality index and the historical coking process data set by using a KAN neural network include: Screening out historical coking process data samples from the historical coking process data set; the historical coking process data samples include coking process condition parameters corresponding to the blended coal produced each time and historical coke quality indicators; A robust normalization method is used to scale the blended coal quality index and coking process condition parameters to obtain a model feature sample set; Establishing a coke quality prediction data set based on the model feature sample set and the historical coke quality indicators; A coke quality prediction model is established using the coke quality prediction data set and the KAN neural network.

2. The precise coking coal blending method according to claim 1, characterized in that: The indicators of a single type of coking coal include: maximum thickness of the gelatinous layer, bonding index, coal rock indicators of a single type of coal and work point data.

3. The precise coking coal blending method according to claim 1, characterized in that: The input variables of the coke quality prediction model include cohesiveness index, volatile matter, moisture, ash content, sulfur content, coking time, coke side temperature and ash composition; the output of the coke quality prediction model includes the ash content, sulfur content, crushing strength, abrasion resistance, reactivity index and post-reaction strength of the coke.

4. The precise coking coal blending method according to claim 1, characterized in that: The steps to obtain the optimal coal blending solution include: Taking the inventory of each type of coal and the sum of all proportions as one as constraints, a sampling proportion optimization model is established to obtain candidate proportion schemes; Inputting the candidate blending scheme, single coal quality parameters, and coking process parameters into the blended coal quality prediction model and the coke quality prediction model to screen out feasible solutions that meet the blended coal quality constraints and the coke quality constraints; The price of the feasible solution is calculated, and the optimal coal blending solution is selected with the lowest blending coal cost as the objective function.

5. The precise coking coal blending method according to claim 4, characterized in that: The calculation formula of the objective function is: ； Among them, n is the total amount of single type of coal used in coal blending; For the The cost price of a single type of coal; For the The ratio of a single type of coal.

6. The precise coking coal blending method according to claim 5, characterized in that: The calculation formula of the sampling ratio optimization model includes: ； Where n is the total amount of single type of coal used in coal blending; C is the cost price of each single type of coal; The proportion of each single type of coal; W To match coal quality indicators; V It is the quality index of coke; For the The proportion of a single type of coal; h j For the The inventory ratio of a single type of coal.

7. The precise coking coal blending method according to claim 4, characterized in that: The quality constraints of the blended coal include the bonding index, gelatinous layer thickness and volatile matter of the blended coal; the quality constraints of the coke include ash content, sulfur content, crushing strength, wear resistance, reactivity index and post-reaction strength.

8. A precise coking coal blending system, characterized in that: The method for precise coking coal blending according to any one of claims 1 to 7 is applicable, wherein the system comprises: A resource information database construction module is used to obtain the indicators of a single type of coking coal to establish a single type of coking coal resource information database; A blending coal quality prediction model building module is used to establish a blending coal quality prediction model based on the single coking coal resource information database and according to a linear model and an SVR fusion model to obtain blending coal quality indicators; A coke quality prediction model building module is used to establish a coke quality prediction model based on the blend coal quality index and the historical coking process data set and through a KAN neural network to obtain the coke quality index; The proportion optimization module is used to establish a sampling proportion optimization model based on the quality parameters of a single coal and the coking process parameters, and with the expected blending coal quality indicators, expected coke quality indicators and inventory as constraints to obtain candidate proportion schemes; and screen the candidate proportion schemes based on the blending coal quality prediction model, the coke quality prediction model and the blending coal cost to obtain the optimal coal blending scheme.

Images