Coking intelligent coal blending multi-scheme optimization method

Abstract

The invention relates to the technical field of coal blending, and discloses a coking intelligent coal blending multi-scheme optimization method, which comprises the following steps: collecting coking coal indexes, blended coal indexes, small coke oven test data, coke oven production data and coal blending scheme data; processing and processing the collected coal blending data, and constructing a self-learning coke quality prediction model; by setting various constraint conditions, the priority of the constraint conditions is adjusted, and a big data optimization algorithm is utilized; based on the coal blending scheme, after the scheme is submitted, and the small coke oven test verification and auditing are carried out, a coal blending list is issued to a silo basic automation PLC control system, and coal preparation production is executed. A coke quality self-learning prediction model under multi-modal data fusion and a multi-objective optimization algorithm aiming at coking production characteristics are adopted, so that a better coal blending scheme is realized, the coal blending cost is reduced, the accuracy of coke quality prediction is improved, coal quality fluctuation is greatly overcome, the production efficiency is improved, and human intervention is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of coal blending, and in particular to a multi-scheme optimization method for intelligent coal blending for coking. Background Art

[0002] Coke is an essential building block for industries like metallurgy and chemical engineering, and its quality has a direct impact on downstream processes and product performance. Traditional coal blending methods rely primarily on manual experience and fixed ratios, lacking the ability to systematically analyze and dynamically adjust the complex relationships between coal characteristics, coke quality, and cost. This makes it difficult to cope with fluctuations in coal sources, market prices, and the increasing demand for coke quality stability.

[0003] With the increasing diversity of coal varieties and the increasing complexity of coal blending targets, how to ensure coke quality while reducing production costs has become a key issue that coking enterprises urgently need to address. In recent years, the rapid development of artificial intelligence, big data analysis, and automated control technologies has provided new technical paths for coking coal blending. Preliminary studies have attempted to use methods such as neural networks and genetic algorithms for coke quality prediction or single-objective optimization. However, most systems only support static solution generation and lack the ability to comprehensively optimize multiple constraints, multiple objectives, and multiple solutions. Furthermore, their low integration with production systems makes it difficult to quickly verify and implement solutions.

[0004] Therefore, it is urgent to propose a multi-scheme optimization method for intelligent coal blending in coking that integrates machine learning prediction, intelligent optimization algorithm and automatic control, realize the intelligent linkage between coke quality prediction and coal blending strategy, improve the scientific nature, execution efficiency and production flexibility of coal blending, and thus better adapt to the development needs of intelligent and efficient modern coking industry. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides a multi-scheme optimization method for intelligent coal blending in coking, which solves the problems of traditional coal blending methods, such as fixed coal blending models, inability to adapt to coal sources, inability to conduct autonomous learning and optimization models when coal quality fluctuates, and limited production efficiency.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a multi-scheme optimization method for intelligent coal blending in coking, comprising the following steps: Collect coking coal indicators, blending coal indicators, small coke oven test data, coke oven production data and coal blending plan data; Processing and handling the collected coal blending data, using a combination of supervised and unsupervised machine learning algorithms to build a self-learning coke quality prediction model; By setting multiple constraints, adjusting the priorities of the constraints, combining the self-learning coke quality prediction model, and using the big data optimization algorithm, multiple groups of coal blending plans are comprehensively optimized and calculated; Based on the coal blending plan, after the plan is submitted, verified by small coke oven test and reviewed, the coal blending order will be issued to the silo basic automation PLC control system to execute coal preparation production.

[0007] Preferably, the collected coking coal indices include: coking coal industrial indices, coal rock indices, cohesiveness indices, and fluidity indices; The blended coal indicators include: ash content, volatile matter, sulfur content, bonding index, and maximum thickness of the gelatinous layer; The test data of the small coke oven include: coke ash content, sulfur content, volatile content, reactivity, post-reaction strength, crushing strength and abrasion resistance; The coke oven production data: coking time, oven temperature; The coal blending scheme data includes: the proportions of various coking coals.

[0008] Preferably, the processing and handling of the collected coal blending data includes: associating the coking coal data corresponding to the coal blending scheme, the blended coal data after blending, the production operation data of the blended coal being loaded into the coke oven for smelting, and the coke data obtained after production, and constructing data samples corresponding to the coking coal data, blended coal data, coke quality data and coke oven production data, the coking coal data including coal type, blending ratio, and coking coal indicators, the coal data including ash content, volatile matter, sulfur content, bonding index, and maximum thickness of the gelatinous layer, the production operation data including coking time and furnace temperature, and the coke data including ash content, sulfur content, reactivity, post-reaction strength, crushing strength and wear resistance.

[0009] Preferably, the step of constructing a self-learning coke quality prediction model using a combined supervised and unsupervised machine learning algorithm comprises: Collect the coking coal data, blending coal data, coke quality data and coke oven production data corresponding to the historical coal blending plan, filter and denoise the collected historical data, and normalize the historical data to organize the historical data into samples for neural network model learning. The normalization formula is as follows: P n =2(PP min ) / (P max -P min )-1; Where: P is the collected original input data; P min 、P max are the maximum and minimum values in P respectively; P n is the input data of the neural network after normalization; The neural network structure is designed to include an input layer, a hidden layer, and an output layer. The input is the ash content, volatile matter, sulfur content, bonding index, maximum thickness of the gel layer, coking time, and furnace temperature of the blended coal. The output is the ash content, sulfur content, post-reaction strength, and crushing strength of the coke. The initial weights and thresholds of the neural network are also set. Input sample data to train the neural network. If the output layer obtains the expected output result, the learning algorithm ends; otherwise, it switches to back propagation, calculates the error signal in reverse according to the original connection path, and adjusts the weights and thresholds of the neurons in each layer through the gradient descent method to minimize the error signal until the error or the number of training times reaches the requirement. Taking the crushing strength of coke as an example, the input layer, hidden layer and output layer use 6, 12 and 1 neurons respectively. The relationship between input and output is as follows: Where: are the weights from the 6 neurons in the input layer to the i-th neuron in the middle layer; is the threshold of the i-th neuron in the middle layer; is the weight from the i-th neuron in the middle layer to the neuron in the output layer; b O1 is the threshold of the output layer neuron; tansig(*) is the extended Sigmoid function; As coke oven production progresses, data samples are continuously accumulated, and the quality indicators in the neural network model learning samples are regularly used to train the coke quality prediction model for model iterative updates and optimization.

[0010] Preferably, the step of comprehensively optimizing and calculating multiple sets of coal blending schemes by setting multiple constraints, adjusting the priorities of the constraints, combining the self-learning coke quality prediction model, and utilizing a big data optimization algorithm includes: setting main coal system ratio restrictions, coking coal inventory constraints, coke quality constraints, and coal blending cost target constraints, as expressed by the following formula: Coal blending cost target constraints: Constraints: x imin ≤x i ≤x imax ; Ad min ≤f(Ad)≤Ad max ; St min ≤f(St)≤St max ; CSR min ≤f(CSR)≤CSR max ; M40 min ≤f(M40)≤M40 max ; Where: f(x) is the coal blending cost, yuan / ton; c i is the current market price of a single type of coal, yuan / ton; x i is the coal type ratio, %; f(x) min 、f(x) max are the upper and lower limits of coal blending cost targets respectively; x imin 、x imax are the upper and lower limits of coal ratio, respectively; f(Ad) is the ash content of coke; Ad max 、Ad min are the upper and lower limits of coke ash content, respectively; f(St) is the coke sulfur content; St max 、St min are the upper and lower limits of coke sulfur content, respectively; f(CSR) is the strength of coke after reaction; CSR max 、CSR min are the upper and lower limits of the strength of coke after reaction, respectively; f(M40) is the crushing strength of coke; M40 max 、M40 min are the upper and lower limits of coke crushing strength, respectively, in %.

[0011] Preferably, the priorities of the main coal system ratio restriction, coke quality constraint, and coal blending cost constraint are adjusted. When optimizing and calculating the coal blending plan, the constraints that cannot be met are given priority and adjusted accordingly. Select the optimization method of optimal cost / optimal quality / optimal comprehensive optimization. Among them, the optimal cost takes the minimum coal blending cost as the objective function, the optimal quality scores the coke quality indicators such as ash content, sulfur content, CSR, CRI, M40, M10, etc., and increases the weight of the score, and takes the maximum coke quality score as the objective function. The optimal comprehensive optimization takes the coal blending cost into consideration on the basis of optimal quality, and takes the maximum comprehensive score as the objective function.

[0012] Preferably, based on the set multiple constraints, adjusting the priority of the constraints, selecting different optimization methods, combining the self-learning coke quality prediction model, using a big data optimization algorithm, iteratively optimizing the proportion population, obtaining the proportion of coal types, and predicting the coke quality, wherein the big data optimization algorithm includes one or more of a differential evolution algorithm, a genetic algorithm, and a simulated annealing algorithm, and the specific steps are as follows: S1 initializes parameters such as population size NP, variation factor F and crossover probability CR, randomly initializes the proportion of each coal type according to the proportion range corresponding to each coal type, and sets the maximum number of iterations g max , get the initialized individuals and generate the g-generation population. The formula is as follows: X k,g ={x k1,g , xk2,g ,…,x kn,g}; Where: k = 1, 2, ..., NP; g is the number of iterations; S2 calculates the target value function of each individual based on the target function of coal blending cost and compares them to find the optimal function value and the optimal individual x gbest,g ; S3 determines whether the maximum number of iterations has been reached, and if so, outputs the result; otherwise, proceeds to the next step; S4 executes steps S5-S8 for each individual in the population, generates NP g+1th generation new individuals, sets g=g+1, and returns to S2; S5 is the X of the population in the g generation k,g Perform mutation operations between individuals to generate mutation intermediates v k,g+1 , the formula is as follows: Where: r1~r4 are 1, 2, ..., any integers that are not equal to each other in NP; S6 will mutate the intermediate v k,g+1 and X in the g-th generation population k,g The individuals are cross-operated to generate the test individual u k,g+1 ={u k1,g+1 ,u k2,g+1 ,…,u kn,g+1}; S7 judges the test individual u k,g+1 Whether each individual in satisfies the constraint condition, if so, execute S8, otherwise adjust u according to the deviation of coke quality index k,g+1 The coal blending ratio of some individuals in the process is calculated and S8 is executed after the constraint conditions are met; S8 judges the test individual u k,g+1 Is the objective function value in better than X? k,g The objective function value of the individual in the experiment is selected k,g+1 Replace X k,g individuals in After multiple iterative calculations, S9 finally obtains the optimal coal blending ratio and coal blending cost target value.

[0013] A coking intelligent coal blending multi-scheme optimization system includes: a coal and coke resource library module for obtaining coking coal indicators, blending coal indicators, small coke oven test data, coke oven production data, and coal blending scheme data; The coke quality prediction module is used to process and process the collected coal blending data and build a self-learning coke quality prediction model using a combination of supervised and unsupervised machine learning algorithms; The coal blending optimization module is used to set multiple constraints, adjust the priority of the constraints, use the big data optimization algorithm, and combine the self-learning coke quality prediction model to comprehensively calculate multiple coal blending plans; The coal preparation guidance module is used to issue the coal preparation order to the silo basic automation PLC control system to execute coal preparation production based on the coal blending plan. After the plan is submitted, verified and reviewed by the small coke oven test, the coal blending order will be issued to the silo basic automation PLC control system.

[0014] The present invention provides a multi-scheme optimization method for intelligent coal blending in coking. It has the following beneficial effects: 1. The present invention adopts a coke quality self-learning prediction model under multimodal data fusion and a multi-objective optimization algorithm based on coking production characteristics to solve the balance problem among constraints such as cost, quality and resources, achieve a better coal blending plan, reduce coal blending costs, improve the accuracy of coke quality prediction, significantly overcome coal quality fluctuations, improve production efficiency, reduce human intervention, and thus improve production stability and reliability.

[0015] 2. The present invention links the optimized coal blending plan with the automated PLC system to achieve closed-loop control from plan generation to on-site coal preparation, resulting in a significant improvement in coal blending execution efficiency and automation level.

[0016] 3. The present invention realizes flexible switching and intelligent scheduling of complex production objectives such as quality priority and cost priority by setting a constraint priority control mechanism, so that the system optimization results are highly consistent with actual production needs.

[0017] 4. The present invention achieves a comprehensive trade-off between quality, cost, and resource constraints by introducing a multi-objective big data optimization algorithm and combining it with the output of a prediction model, thereby automatically generating multiple sets of high-quality, low-cost coal blending solutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a method flow chart of a multi-scheme optimization method for intelligent coal blending for coking according to the present invention; Figure 2 This is a system architecture diagram of a coking intelligent coal blending multi-scheme optimization system of the present invention. DETAILED DESCRIPTION

[0019] The following will clearly and completely describe the technical solution of the present invention in conjunction with the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0020] Please see the attached Figure 1The embodiment of the present invention provides a multi-scheme optimization method for intelligent coal blending in a coking plant, comprising the following steps: Collect coking coal indicators, blending coal indicators, small coke oven test data, coke oven production data and coal blending plan data; Processing and handling the collected coal blending data, using a combination of supervised and unsupervised machine learning algorithms to build a self-learning coke quality prediction model; By setting multiple constraints, adjusting the priority of the constraints, combining the self-learning coke quality prediction model, and using the big data optimization algorithm, multiple sets of coal blending plans are calculated through comprehensive optimization. Based on the coal blending plan, after the plan is submitted, verified and reviewed by the small coke oven test, the coal blending order will be issued to the silo basic automation PLC control system to execute coal preparation production.

[0021] The collection of coking coal indicators includes: coking coal industrial indicators, coal rock indicators, cohesiveness indicators, and fluidity indicators; The coal blending indicators include: ash content, volatile matter, sulfur content, bonding index, and maximum thickness of the colloidal layer; Small coke oven test data include: coke ash content, sulfur content, volatile matter, reactivity, post-reaction strength, crushing strength and abrasion resistance; Coke oven production data: coking time, oven temperature; Coal blending plan data includes: the proportion of each coking coal.

[0022] Specifically, data is first collected from five dimensions: coking coal indicators, blending coal indicators, small coke oven test data, coke oven production data, and coal blending plan data. These data cover the basic characteristics of coal types, operating parameters of the production process, and quality indicators of the final coke, ensuring the comprehensiveness and accuracy of the data, and providing strong support for subsequent model training; in the data processing stage, the present invention combines supervised and unsupervised learning methods to construct a self-learning coke quality prediction model. The self-learning mechanism optimizes the model through dynamic updates to ensure that the prediction accuracy continues to improve with the accumulation of new data. By filtering, denoising, and normalizing the data, a neural network model that adapts to different coal types and production conditions is constructed. The model can predict coke quality and support dynamic adjustment; Based on the coke quality prediction model, the present invention uses a big data optimization algorithm and combines multiple constraints (such as proportion, inventory, coke quality, and cost constraints) to perform multi-objective optimization. The innovation lies in setting the priority of the constraints, enabling the system to find the optimal balance between multiple objectives. In addition, using multiple optimization methods such as differential evolution algorithm, genetic algorithm, and simulated annealing algorithm, the system can efficiently calculate multiple feasible coal blending schemes to meet different production needs. By dynamically adjusting constraint priorities and using different optimization methods (such as cost priority, quality priority, etc.), the present invention can flexibly respond to different production environments and optimize the production process.

[0023] The collected coal blending data are processed and handled, including: associating the coking coal data corresponding to the coal blending plan, the blended coal data after blending, the production operation data of the blended coal being loaded into the coke oven for smelting, and the coke data obtained after production, and constructing data samples corresponding to the coking coal data, blended coal data, coke quality data and coke oven production data. The coking coal data includes coal type, blending ratio, and coking coal indicators. The coal data includes ash content, volatile matter, sulfur content, bonding index, and maximum thickness of the gelatinous layer. The production operation data includes coking time and furnace temperature. The coke data includes ash content, sulfur content, reactivity, post-reaction strength, crushing strength, and wear resistance.

[0024] Specifically, the processing and handling of coal blending data involves associating the data of various coking coals, blending coals, production operation data, and post-production coke quality data to form a complete sample library, providing a sufficient training basis for the subsequent coke quality prediction model; During the data integration process, the characteristic data of each coking coal in the coal blending plan is first combined with the data of the blended coal. Coking coal data includes coal type, coal ratio, and chemical indicators of each type of coking coal, such as ash content, volatile matter, sulfur content, caking index, and maximum thickness of the colloidal layer. This data directly affects the coking characteristics of the coal type and is a key input for building the prediction model. At the same time, the integration of coal data, including ash content, volatile matter, sulfur content, caking index and maximum thickness of the colloidal layer, reflects the impact of different coal types on coke quality after mixing. The changes in these data after the coal type ratio plays an important role in the final quality of coke; In addition, production operation data (such as coking time, furnace temperature, etc.) are also closely related to the coal type ratio and coke quality. These data reflect the operating conditions of the coke oven in actual operation and can directly affect the pyrolysis behavior during the coking process, thereby affecting the quality of the coke. Finally, coke quality data (such as ash content, sulfur content, reactivity, post-reaction strength, crushing strength, abrasion resistance, etc.) are combined with other data to form a comprehensive data sample library. These coke quality indicators are crucial in evaluating coking performance and can directly reflect the impact of different coal blending schemes on coke quality. By integrating and correlating these data, the present invention not only more accurately captures the impact of coal type and production conditions on coke quality, but also provides a solid data foundation for building a self-learning coke quality prediction model. This multidimensional data correlation processing method is one of the key innovations of the present invention in the data processing link, effectively improving the accuracy and practicality of the prediction model.

[0025] The steps for building a self-learning coke quality prediction model using a combination of supervised and unsupervised machine learning algorithms include: Collect the coking coal data, blending coal data, coke quality data and coke oven production data corresponding to the historical coal blending plan, filter and denoise the collected historical data, and normalize the historical data to organize the historical data into samples for neural network model learning. The normalization formula is as follows: P n =2(PP min ) / (P max -P min )-1; Where: P is the collected original input data; P min 、P max are the maximum and minimum values in P respectively; P n is the input data of the neural network after normalization; The neural network structure is designed to include an input layer, a hidden layer, and an output layer. The input is the ash content, volatile matter, sulfur content, bonding index, maximum thickness of the gel layer, coking time, and furnace temperature of the blended coal. The output is the ash content, sulfur content, post-reaction strength, and crushing strength of the coke. The initial weights and thresholds of the neural network are also set. Input sample data to train the neural network. If the output layer obtains the expected output result, the learning algorithm ends; otherwise, it switches to back propagation, calculates the error signal in reverse according to the original connection path, and adjusts the weights and thresholds of the neurons in each layer through the gradient descent method to minimize the error signal until the error or the number of training times reaches the requirement. Taking the crushing strength of coke as an example, the input layer, hidden layer and output layer use 6, 12 and 1 neurons respectively. The relationship between input and output is as follows: Where: are the weights from the 6 neurons in the input layer to the i-th neuron in the middle layer; is the threshold of the i-th neuron in the middle layer; is the weight from the i-th neuron in the middle layer to the neuron in the output layer; b O1 is the threshold of the output layer neuron; tansig(*) is the extended Sigmoid function; As coke oven production progresses, data samples are continuously accumulated, and the quality indicators in the neural network model learning samples are regularly used to train the coke quality prediction model for model iterative updates and optimization.

[0026] Specifically, during the data processing phase, we first collected coking coal data, blended coal data, coke quality data, and coke oven production data corresponding to historical coal blending plans. To ensure the stability and efficiency of model training, we first filtered and denoised the collected data to remove outliers and interference signals. Afterwards, all collected raw data was normalized using the following formula: Among them, P is the original input data, P min and P max Represent the minimum and maximum values in the data set, P n This is the normalized data, which will be used as the input of the neural network. Normalization helps to eliminate the scale differences between different features, making the neural network more effective in learning; The neural network design includes an input layer, a hidden layer, and an output layer. The input layer receives multiple characteristic variables, such as the ash content (Ad), volatile matter (Vdaf), sulfur content (St), caking index (G), maximum thickness of the colloidal layer (Ymax), coking time (T), and furnace temperature (Temp) of the blended coal. These input variables directly affect the quality of the coke and therefore must be used as input to the model. The output layer predicts the quality indicators of the coke, such as ash content (Ad), sulfur content (St), reactivity (CRI), strength after reaction (CSR), crushing strength (M40), etc. The training process of a neural network includes initializing weights and thresholds, and training the network by inputting sample data. The training uses the backpropagation algorithm. During this process, if the output layer's results meet the desired goal, the learning process ends; otherwise, the error signal is backpropagated, and the weights and thresholds of each layer of neurons are adjusted by the gradient descent method until the error is minimized or the predetermined number of training times is reached. The hierarchical structure of a neural network is as follows: The input layer contains 6 neurons, corresponding to 6 input features; The hidden layer contains 12 neurons, which are used to perform nonlinear mapping on the input data; The output layer contains 1 neuron, which outputs the quality prediction results of coke; As coke oven production continues, data samples continuously accumulate. Therefore, the present invention utilizes a self-learning mechanism, enabling the coke quality prediction model to undergo regular iterative updates and optimizations after each new data collection. Each addition of new data retrains the neural network model, further improving its prediction accuracy. This continuous model training and optimization process ensures the real-time and accurate prediction of coke quality, adapting to various changes that may occur during the production process.

[0027] By setting multiple constraints, adjusting the priority of the constraints, combining the self-learning coke quality prediction model, and using the big data optimization algorithm, the steps of comprehensively optimizing and calculating multiple groups of coal blending plans include: Set the main coal system ratio limit, coking coal inventory constraint, coke quality constraint and coal blending cost target constraint. The formula is as follows: Coal blending cost target constraints: Constraints: x imin ≤x i ≤x imax ; Ad min ≤f(Ad)≤Ad max ; St min ≤f(St)≤St max ; CSR min ≤f(CSR)≤CSR max ; M40 min ≤f(M40)≤M40 max ; Where: f(x) is the coal blending cost, yuan / ton; c i is the current market price of a single type of coal, yuan / ton; x i is the coal type ratio, %; f(x) min 、f(x) max are the upper and lower limits of coal blending cost targets respectively; x imin 、x imax are the upper and lower limits of coal ratio, respectively; f(Ad) is the ash content of coke; Ad max 、Ad min are the upper and lower limits of coke ash content, respectively; f(St) is the coke sulfur content; St max 、St min are the upper and lower limits of coke sulfur content, respectively; f(CSR) is the strength of coke after reaction; CSR max 、CSR min are the upper and lower limits of the strength of coke after reaction, respectively; f(M40) is the crushing strength of coke; M40 max 、M40 min are the upper and lower limits of coke crushing strength, respectively, in %.

[0028] Specifically, the setting of the coal blending cost target constraint is intended to ensure the cost feasibility of the selected coal blending scheme. The coal blending cost target is calculated based on the market price of each coal and its blending ratio. Specifically, the coal blending cost target constraint formula is as follows: f(x)min ≤f(x)≤f(x) max ; Among them, f(x) represents the coal blending cost (yuan / ton), c i is the market price of a single type of coal (yuan / ton), x i is the coal type ratio (%), f(x) min and f(x) max They are the upper and lower limits of coal blending cost; Coke quality is one of the key constraints in optimizing coal blending. Key coke quality indicators include ash content (Ad), sulfur content (St), carbon steel strength after reaction (CSR), and crushing strength (M40). Each indicator must meet specific quality ranges, with the following constraints: Coke ash constraint: f(Ad) min ≤f(Ad)≤f(Ad) max ; Coke sulfur constraint: f(St) min ≤f(St)≤f(St) max ; Coke strength constraint after reaction (CSR): f(CSR) min ≤f(CSR)≤f(CSR) max ; Coke crushing strength constraint (M40): f(M40) min ≤f(M40)≤f(M40) max ; Among them, the upper and lower limits of each quality index (such as f(Ad) min ,f(Ad) max ) is set according to the standards and requirements of coke production to ensure that the resulting coal blending plan meets the requirements in terms of coke quality; Coal blending is also an important constraint in the optimization process. The blending ratio of each type of coal must be within a reasonable range, and the sum of the blending ratios of all coal types should be 100%. The constraint formula is as follows: x min ≤x i ≤x max , Among them, x i is the ratio of each coal, and each x i The values are within the specified upper and lower limits; These constraints are comprehensively calculated using an optimization algorithm to ensure that the resulting coal blending solution meets production requirements while minimizing costs and optimizing coke quality. By continuously adjusting the priorities of these constraints, the system can flexibly adapt to different production needs and select the optimal coal blending solution.

[0029] Adjust the priority of main coal system proportion restrictions, coke quality constraints, and coal blending cost constraints. When optimizing and calculating coal blending plans, prioritize unsatisfactory constraints and make adjustments accordingly. Select the optimization method of optimal cost / optimal quality / optimal comprehensive optimization. Among them, the optimal cost takes the minimum coal blending cost as the objective function, the optimal quality scores the coke quality indicators such as ash content, sulfur content, CSR, CRI, M40, M10, etc., and increases the weight of the score, and takes the maximum coke quality score as the objective function. The optimal comprehensive optimization takes the coal blending cost into consideration on the basis of optimal quality, and takes the maximum comprehensive score as the objective function.

[0030] Specifically, the system adjusts the priority of constraints through the following mechanisms: Main coal system ratio constraints: Ratio constraints ensure that the ratio of each coal is within an acceptable range and the total is 100%. During the optimization process, when the ratio does not meet the requirements, the system will immediately prompt and prioritize adjustments to avoid unreasonable coal allocation.

[0031] Coke quality constraints: Coke quality indicators (such as ash, sulfur, reactivity, and crushing strength) are key constraints for coal blending. During the optimization process, if coke quality indicators do not meet requirements, the system prioritizes adjustments to these indicators. The system scores key coke quality indicators (such as CSR, CRI, and M40) and adjusts optimization directions based on these goals.

[0032] Coal blending cost constraints: Cost constraints ensure that the cost of the coal blending plan meets the budget. When the coal blending cost fails to meet the requirements, the system will automatically prompt and adjust the coal type and proportion in the coal blending plan to optimize the coal blending cost; The system provides three different optimization methods, and users can choose different optimization objectives according to their specific needs. The objective function and constraints of each optimization method are different, as follows: Objective function: Minimizing coal blending cost. This method takes coal blending cost as the target and prioritizes finding a coal blending solution that minimizes the cost. Under the premise of meeting the coke quality and blending requirements, the total cost is minimized by adjusting the coal type ratio, raw material selection, and production conditions. The formula is as follows: Among them, c i is the unit price of the i-th type of coal, x i It is the coal ratio.

[0033] Objective function: Maximize the coke quality score. This method prioritizes the quality of coke by scoring the quality indicators of coke (such as ash content, sulfur content, reactivity, crushing strength, etc.). In the quality optimization scheme, the scoring is based on the following: S total=w1·S Ad +w2·S St +w3·S CSR +w4·S M40 +…; Among them, w1, w2, w3, w4, ... are the weights of each indicator, S Ad ,S St ,S CSR ,S M40 ,…are the scores of various coke quality indicators. In the quality optimization method, the maximum value of the coke quality score is taken as the target. The system maximizes the comprehensive quality score of coke by adjusting parameters such as coal type ratio and operating conditions; Objective function: Comprehensively consider cost and quality to seek the best comprehensive score. The comprehensive optimal method takes into account the cost of coal blending on the basis of optimal quality. The comprehensive score is calculated by the following formula: S composite =αS quality +(1-α)·S cost ; Among them, S quality is the quality score, S cost is the cost score, α is the weight coefficient (0≤α≤1), which controls the balance between quality and cost; These optimization methods offer flexible options, allowing users to choose optimal cost, optimal quality, or a combination of these, tailored to their production requirements. Within each optimization method, the system generates the most appropriate coal blending solution based on the adjusted constraints and objective function, ensuring an optimal balance between coke quality, production costs, and coal type ratios.

[0034] Based on the set multiple constraints, adjustment of the priority of the constraints, selection of different optimization methods, combined with the self-learning coke quality prediction model, the big data optimization algorithm is used to iteratively optimize the proportion population, obtain the proportion of coal types, and predict the coke quality. Among them, the big data optimization algorithm includes one or more of the differential evolution algorithm, genetic algorithm, and simulated annealing algorithm. The specific steps are as follows: S1 initializes parameters such as population size NP, variation factor F and crossover probability CR, randomly initializes the proportion of each coal type according to the proportion range corresponding to each coal type, and sets the maximum number of iterations g max , get the initialized individuals and generate the g-generation population. The formula is as follows: X k,g ={x k1,g , x k2,g ,…,x kn,g}; Where: k = 1, 2, ..., NP; g is the number of iterations; S2 calculates the target value function of each individual based on the target function of coal blending cost and compares them to find the optimal function value and the optimal individual x gbest,g ; S3 determines whether the maximum number of iterations has been reached, and if so, outputs the result; otherwise, proceeds to the next step; S4 executes steps S5-S8 for each individual in the population, generates NP g+1th generation new individuals, sets g=g+1, and returns to S2; S5 is the X of the population in the g generation k,g Perform mutation operations between individuals to generate mutation intermediates v k,g+1 , the formula is as follows: Where: r1~r4 are 1, 2, ..., any integers that are not equal to each other in NP; S6 will mutate the intermediate v k,g+1 and X in the g-th generation population k,g The individuals are cross-operated to generate the test individual u k,g+1 ={u k1,g+1 ,u k2,g+1 ,…,u kn,g+1}; S7 judges the test individual u k,g+1 Whether each individual in satisfies the constraint condition, if so, execute S8, otherwise adjust u according to the deviation of coke quality index k,g+1 The coal blending ratio of some individuals in the process is calculated and S8 is executed after the constraint conditions are met; S8 judges the test individual u k,g+1 Is the objective function value in better than X? k,g The objective function value of the individual in the experiment is selected k,g+1 Replace X k,g individuals in After multiple iterative calculations, S9 finally obtains the optimal coal blending ratio and coal blending cost target value.

[0035] Specifically, the coal blending scheme is optimized using big data optimization algorithms (such as differential evolution, genetic algorithms, or simulated annealing) by combining the set multiple constraints, adjusting the constraint priorities, and different optimization methods. First, the population is initialized and the target value of each individual is calculated based on the coal blending cost objective function. By judging whether the maximum number of iterations has been reached, if not, the subsequent steps are continued. Then, the individuals in the population are mutated and crossover operations are performed to generate a new generation of individuals, and whether the constraints are met is checked. If not, the individual ratio is adjusted according to the coke quality deviation. Finally, through multiple iterations, the optimal coal blending scheme is selected to ensure the best balance between coke quality and cost.

[0036] Please see the attached Figure 2,A coking intelligent coal blending multi-scheme optimization system includes: a coal and coke resource library module for obtaining coking coal indicators, blending coal indicators, small coke oven test data, coke oven production data, and coal blending scheme data; The coke quality prediction module is used to process and process the collected coal blending data and build a self-learning coke quality prediction model using a combination of supervised and unsupervised machine learning algorithms; The coal blending optimization module is used to set multiple constraints, adjust the priority of the constraints, and use the big data optimization algorithm combined with the self-learning coke quality prediction model to comprehensively calculate multiple coal blending plans; The coal preparation guidance module is used for coal blending scheme based on the actual situation. After the scheme is submitted, verified by small coke oven test and reviewed, the coal blending order will be issued to the silo basic automation PLC control system to execute coal preparation production.

[0037] Specifically, the coal and coke resource library module ensures the comprehensiveness and accuracy of data by centrally managing the characteristic data of various coal types, production operation data, and historical coke quality data. It provides reliable data support for subsequent coke quality prediction and coal blending plan optimization. The coke quality prediction module uses a dynamically updated prediction model, allowing the system to continuously improve the accuracy of coke quality predictions as new data is added. It provides accurate quality predictions for coal blending optimization, helping to avoid coal blending combinations that do not meet quality standards, thereby reducing production risks. The coal blending optimization module can calculate multiple optimization schemes by integrating multiple constraints and select the best scheme under the premise of ensuring the balance between quality and cost. Through intelligent optimization, it can reduce production costs, improve coke quality and optimize production efficiency. The Coal Preparation Guidance Module ensures the accurate and smooth execution of optimized coal blending plans, reducing manual intervention and increasing production automation. Tests have proven that the implementation of optimized plans can significantly improve coke quality and reduce production costs.

[0038] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A multi-scheme optimization method for intelligent coal blending in coking, characterized in that: The following steps are involved: Collect coking coal indicators, blending coal indicators, small coke oven test data, coke oven production data and coal blending plan data; Processing and handling the collected coal blending data, using a combination of supervised and unsupervised machine learning algorithms to build a self-learning coke quality prediction model; By setting multiple constraints, adjusting the priorities of the constraints, combining the self-learning coke quality prediction model, and using the big data optimization algorithm, multiple groups of coal blending plans are comprehensively optimized and calculated; Based on the coal blending plan, after the plan is submitted, verified by small coke oven test and reviewed, the coal blending order will be issued to the silo basic automation PLC control system to execute coal preparation production.

2. The multi-scheme optimization method for intelligent coal blending in coking according to claim 1, characterized in that: The collected coking coal indices include: coking coal industrial indices, coal rock indices, cohesiveness indices, and fluidity indices; The blended coal indicators include: ash content, volatile matter, sulfur content, bonding index, and maximum thickness of the gelatinous layer; The test data of the small coke oven include: coke ash content, sulfur content, volatile content, reactivity, post-reaction strength, crushing strength and abrasion resistance; The coke oven production data: coking time, oven temperature; The coal blending scheme data includes: the proportions of various coking coals.

3. The multi-scheme optimization method for intelligent coal blending in coking according to claim 1, characterized in that: The processing and handling of the collected coal blending data includes: associating the coking coal data corresponding to the coal blending scheme, the blended coal data after blending, the production operation data of the blended coal being loaded into the coke oven for smelting, and the coke data obtained after production, and constructing data samples corresponding to the coking coal data, the blended coal data, the coke quality data and the coke oven production data. The coking coal data includes coal type, blending ratio, and coking coal indicators. The coal data includes ash content, volatile matter, sulfur content, bonding index, and maximum thickness of the gelatinous layer. The production operation data includes coking time and furnace temperature. The coke data includes ash content, sulfur content, reactivity, post-reaction strength, crushing strength and wear resistance.

4. The multi-scheme optimization method for intelligent coal blending in coking according to claim 1, characterized in that: The step of using a combined supervised and unsupervised machine learning algorithm to construct a self-learning coke quality prediction model includes: Collect the coking coal data, blending coal data, coke quality data and coke oven production data corresponding to the historical coal blending plan, filter and denoise the collected historical data, and normalize the historical data to organize the historical data into samples for neural network model learning. The normalization formula is as follows: P n =2(PP min ) / (P max -P min )-1; Where: P is the collected original input data; P min 、P max are the maximum and minimum values in P respectively; P n is the input data of the neural network after normalization; The neural network structure is designed to include an input layer, a hidden layer, and an output layer. The input is the ash content, volatile matter, sulfur content, bonding index, maximum thickness of the gel layer, coking time, and furnace temperature of the blended coal. The output is the ash content, sulfur content, post-reaction strength, and crushing strength of the coke. The initial weights and thresholds of the neural network are also set. Input sample data to train the neural network. If the output layer obtains the expected output result, the learning algorithm ends; otherwise, it switches to back propagation, calculates the error signal in reverse according to the original connection path, and adjusts the weights and thresholds of the neurons in each layer through the gradient descent method to minimize the error signal until the error or the number of training times reaches the requirement. Taking the crushing strength of coke as an example, the input layer, hidden layer and output layer use 6, 12 and 1 neurons respectively. The relationship between input and output is as follows: Where: are the weights from the 6 neurons in the input layer to the i-th neuron in the middle layer; is the threshold of the i-th neuron in the middle layer; is the weight from the i-th neuron in the middle layer to the neuron in the output layer; b O1 is the threshold of the output layer neuron; tansig(*) is the extended Sigmoid function; As coke oven production progresses, data samples are continuously accumulated, and the quality indicators in the neural network model learning samples are regularly used to train the coke quality prediction model for model iterative updates and optimization.

5. The multi-scheme optimization method for intelligent coal blending in coking according to claim 1, characterized in that: The steps of setting multiple constraints, adjusting the priorities of the constraints, combining the self-learning coke quality prediction model, and utilizing a big data optimization algorithm to comprehensively optimize and calculate multiple groups of coal blending plans include: Set the main coal system ratio limit, coking coal inventory constraint, coke quality constraint and coal blending cost target constraint. The formula is as follows: Coal blending cost target constraints: Constraints: x imin ≤x i ≤x imax ; Ad min ≤f(Ad)≤Ad max ; St min ≤f(St)≤St max ; CSR min ≤f(CSR)≤CSR max ; M40 min ≤f(M40)≤M40 max ; Where: f(x) is the coal blending cost, yuan / ton; c i is the current market price of a single type of coal, yuan / ton; x i is the coal type ratio, %; f(x) min 、f(x) max are the upper and lower limits of coal blending cost targets respectively; x imin 、x imax are the upper and lower limits of coal ratio, respectively; f(Ad) is the ash content of coke; Ad max 、Ad min are the upper and lower limits of coke ash content, respectively; f(St) is the coke sulfur content; St max 、St min are the upper and lower limits of coke sulfur content, respectively; f(CSR) is the strength of coke after reaction; CSR max 、CSR min are the upper and lower limits of the strength of coke after reaction, respectively; f(M40) is the crushing strength of coke; M40 max 、M40 min are the upper and lower limits of coke crushing strength, respectively, in %.

6. The multi-scheme optimization method for intelligent coal blending in coking according to claim 1, characterized in that: Adjust the priority of main coal system proportion restrictions, coke quality constraints, and coal blending cost constraints. When optimizing and calculating coal blending plans, prioritize unsatisfactory constraints and make adjustments accordingly. Select the optimization method of optimal cost / optimal quality / optimal comprehensive optimization. Among them, the optimal cost takes the minimum coal blending cost as the objective function, the optimal quality scores the coke quality indicators such as ash content, sulfur content, CSR, CRI, M40, M10, etc., and increases the weight of the score, and takes the maximum coke quality score as the objective function. The optimal comprehensive optimization takes the coal blending cost into consideration on the basis of optimal quality, and takes the maximum comprehensive score as the objective function.

7. The multi-scheme optimization method for intelligent coal blending in coking according to claim 1, characterized in that: Based on the set multiple constraints, adjusting the priority of the constraints, selecting different optimization methods, combining the self-learning coke quality prediction model, and using the big data optimization algorithm, the proportion population is iteratively optimized to obtain the proportion of coal types and predict the coke quality. The big data optimization algorithm includes one or more of a differential evolution algorithm, a genetic algorithm, and a simulated annealing algorithm. The specific steps are as follows: S1 initializes parameters such as population size NP, variation factor F and crossover probability CR, randomly initializes the proportion of each coal type according to the proportion range corresponding to each coal type, and sets the maximum number of iterations g max , get the initialized individuals and generate the g-generation population. The formula is as follows: X k,g ={x k1,g ,x k2,g ,…,x kn,g }; Where: k = 1, 2, ..., NP; g is the number of iterations; S2 calculates the target value function of each individual based on the target function of coal blending cost and compares them to find the optimal function value and the optimal individual x gbest,g ; S3 determines whether the maximum number of iterations has been reached, and if so, outputs the result; otherwise, proceeds to the next step; S4 executes steps S5-S8 for each individual in the population, generates NP g+1th generation new individuals, sets g=g+1, and returns to S2; S5 is the X of the population in the g generation k,g Perform mutation operations between individuals to generate mutation intermediates v k,g+1 , the formula is as follows: Where: r1~r4 are 1, 2, ..., any integers that are not equal to each other in NP; S6 will mutate the intermediate v k,g+1 and X in the g-th generation population k,g The individuals are cross-operated to generate the test individual u k,g+1 ={u k1,g+1 ,u k2,g+1 ,…,u kn,g+1 }; S7 judges the test individual u k,g+1 Whether each individual in satisfies the constraint condition, if so, execute S8, otherwise adjust u according to the deviation of coke quality index k,g+1 The coal blending ratio of some individuals in the test is determined until the constraint conditions are met and S8 is executed; S8 determines the test individual u k,g+1 Is the objective function value in better than X? k,g The objective function value of the individual in the experiment is selected k,g+1 Replace X k,g individuals in After multiple iterative calculations, S9 finally obtains the optimal coal blending ratio and coal blending cost target value.

8. A coking intelligent coal blending multi-scheme optimization system, applied to a coking intelligent coal blending multi-scheme optimization method according to any one of claims 1 to 7, characterized in that: include: The coal and coke resource library module is used to obtain coking coal indicators, blending coal indicators, small coke oven test data, coke oven production data, and coal blending plan data; The coke quality prediction module is used to process and process the collected coal blending data and build a self-learning coke quality prediction model using a combination of supervised and unsupervised machine learning algorithms; The coal blending optimization module is used to set multiple constraints, adjust the priority of the constraints, use the big data optimization algorithm, and combine the self-learning coke quality prediction model to comprehensively calculate multiple coal blending plans; The coal preparation guidance module is used to issue the coal preparation order to the silo basic automation PLC control system to execute coal preparation production based on the coal blending plan. After the plan is submitted, verified and reviewed by the small coke oven test, the coal blending order will be issued to the silo basic automation PLC control system.

Images