Coking coal feeding intelligent prediction method and system

Abstract

The invention relates to an intelligent coking coal feeding prediction method and system, and belongs to the technical field of intelligent coking management. An expert guidance model is applied to deeply analyze feed coal characteristics according to coal quality detection data, a feed coal analysis result is determined, and constraint conditions are determined according to feed coal, coke quality indexes and cost indexes; preliminarily generating a coal blending scheme by applying a mathematical calculation model, adopting a mixed integer rule and a genetic algorithm, taking coke quality index and cost minimization as targets and combining constraint conditions; the coal blending scheme is adjusted based on multiple interactions of the mathematical calculation model and the expert guidance model, and the expert guidance model determines scheme optimization suggestions according to the coal blending scheme; applying a market prediction model based on an LSTM neural network and time sequence causal analysis to predict the price trend of the feed coal, and generating a procurement scheme suggestion corresponding to the coal blending scheme according to the price trend and inventory information of the feed coal; and generating a coal feeding evaluation report. The method has the effect of accurately optimizing the coal blending scheme.

Description

Technical Field

[0001] This application relates to the technical field of intelligent management of coking plants, and in particular to an intelligent prediction method and system for coal feeding into coking plants. Background Technology

[0002] In the coking industry, coking plants use a multi-coal blending method to produce coke. The coal blending process plays a crucial role in cost control and product quality assurance, accounting for approximately 50% of the total production cost. Therefore, accurately controlling coal blending costs is of great significance for cost reduction and efficiency improvement. At the same time, stabilizing coke quality is one of the core tasks of coking enterprises.

[0003] Traditional methods for determining coal input and formulating coal blending schemes in coking plants have many drawbacks. On the one hand, mainstream coal blending theories, such as the principle of overlapping plastic layers, the principle of interactive coal blending, and the principle of co-carbonization, are only applicable to laboratory research and coal blending and coking results, lacking operability in actual coking production. On the other hand, most coking enterprises rely on experience-based coal blending practices. However, once the coal source changes, a large number of coal blending experiments and production practices are needed to re-verify these methods. This not only consumes a lot of manpower, material resources, and time, but also fails to achieve the goal of precise control over coke quality and cost.

[0004] Therefore, there is an urgent need for an intelligent prediction method for coking coal input that can accurately optimize coal blending schemes, which is of great significance for improving the market competitiveness and sustainable development capabilities of coking enterprises. Summary of the Invention

[0005] To accurately optimize coal blending schemes, this application provides a method and system for intelligent prediction of coal feed into coking plants.

[0006] This application provides a smart prediction method for coal feed into coking plants, employing the following technical solution: Obtain quality and cost indicators for various raw coals and cokes used in coal blending; Real-time access to coal quality testing data, market prices, and inventory information for each type of raw coal; The application of the expert-guided model is used to conduct in-depth analysis of the characteristics of the raw coal based on the coal quality test data, determine the analysis results of the raw coal, and determine the constraints based on the raw coal, the coke quality indicators, and the cost indicators. Based on the coal quality testing data, market price, and inventory information of each type of raw coal, a preliminary coal blending scheme is generated by applying a mathematical calculation model using mixed integer rules and genetic algorithms, with the goal of minimizing coke quality indicators and costs, and in conjunction with the constraints. Based on multiple interactions between the mathematical calculation model and the expert guidance model, the coal blending scheme is adjusted, and the expert guidance model determines optimization suggestions based on the coal blending scheme. A market forecasting model based on LSTM neural network and time-series causal analysis is applied to predict the price trend of raw coal in the future within a preset time period. Based on the price trend of raw coal and the inventory information, a procurement plan suggestion corresponding to the coal blending scheme is generated. Generate a coal intake evaluation report that includes at least the raw coal analysis results, the coal blending scheme, the scheme optimization suggestions, and the procurement scheme suggestions.

[0007] By adopting the above technical solution, a full-process coal blending decision-making system is constructed by integrating basic raw coal information, real-time coal quality data, price, and inventory information. First, experts guide the model to analyze coal quality and establish constraints. Then, an initial coal blending plan is generated by combining mathematical models and intelligent algorithms. Next, the plan is optimized through model interaction to ensure that coke quality meets standards while minimizing costs. Simultaneously, market forecasting models predict price trends and, in conjunction with inventory data, generate procurement recommendations, ultimately forming a comprehensive evaluation report. This overall approach achieves scientific and dynamic optimization of the coal blending plan, balancing quality and cost, and reduces supply chain risks through procurement forecasting, thereby improving the intelligence and economy of the entire coal blending chain.

[0008] Furthermore, based on the coal quality testing data, market price, and inventory information for each type of raw coal, a mathematical calculation model employing mixed integer rules and a genetic algorithm is used to generate a coal blending scheme with the goal of minimizing coke quality indicators and costs. This scheme is combined with an expert-guided model based on the coal blending constraints given by the raw coal, including: Construct the objective function using mixed-integer programming; The objective function is: minF=ω1C+ω2Q C=∑ i,t x ijt ·p i ·m t Where C is the coal blending cost, x ijt Let p be the proportion of the i-th type of raw coal in the j-th coal blending scheme during the t-th period. i m is the unit price of the i-th type of raw coal. t Where Q represents the total amount of coal to be blended, and Q is the penalty for coke quality deviation. Let k be the measured value of the k-th quality indicator. Let λ be the target value of the k-th quality indicator. k Let ω1 and ω2 be the weights corresponding to the k-th quality indicator, and ω1 and ω2 be the weight coefficients of coal blending cost and coke quality, respectively. The constraints are as follows: I it=I i,t-1 +y it ·s it -x ijt ·m t ≥I min Among them, a ik Let b be the k-th index value of the i-th type of raw coal. k I represents the upper or lower limit of the value of the k-th indicator. it Let s be the inventory of the i-th type of raw coal in period t. it Let y be the purchase quantity of the i-th type of raw coal in period t. it I is an indicator variable for whether the i-th type of raw coal was purchased. min Safety stock level; Based on the inventory information and coal quality testing data, N feasible coal blending schemes that satisfy the objective function and the constraints are randomly generated. For each feasible coal blending scheme, the proportions and types of raw coal are iteratively optimized using a genetic algorithm, while satisfying the aforementioned constraints. After each iteration, the fitness of the feasible coal blending scheme is calculated based on the market price and inventory information, combined with the objective function. After multiple iterations of optimization, at least one feasible coal blending scheme with a fitness greater than a preset value is determined as the coal blending scheme. By adopting the above technical solution, the objective function and constraints are constructed through mixed integer programming, which can accurately express the complex relationships in the coal blending problem. N feasible schemes that satisfy the constraints are randomly generated, providing a rich initial population for the genetic algorithm, avoiding the algorithm from getting trapped in local optima, and increasing the possibility of finding the global optimum. The genetic algorithm optimizes the coal blending ratio and raw coal type at the same time, which can comprehensively explore the solution space of the coal blending scheme and explore better raw material combinations and ratios. The fitness is calculated in each iteration in combination with real-time market price and inventory information, so that the algorithm can dynamically respond to market fluctuations and ensure the economy and feasibility of the final scheme.

[0009] Furthermore, the application, based on a market forecasting model using LSTM neural networks and time-series causal analysis, predicts the price trend of raw coal over a predetermined time period, including: Causality tests were applied to quantify the causal influence of multiple exogenous variables on raw coal prices; Obtain historical price data for each type of raw coal; The time series features of the historical price data and the causal influence strength of the corresponding exogenous variables are obtained, and the time series features are standardized to obtain a dataset consisting of historical price data, its corresponding time series features, and Granger causal variable data. Construct an LSTM model, train the LSTM model using a dataset, and obtain the trained LSTM market price prediction model. Obtain real-time Granger causal variable data and time series features corresponding to market prices; The market price of the raw coal, the time series characteristics corresponding to the market price, and the real-time Granger causal variable data are input into the trained LSTM market price prediction model to predict the price at multiple future moments within a preset future time period. The price trend is obtained based on the predicted price at each future moment.

[0010] By employing the aforementioned technical solution, this market forecasting technology first quantifies the impact of exogenous variables on raw coal prices through causal testing. It then constructs a dataset based on the time-series characteristics of historical prices, trains an LSTM model, and finally inputs real-time causal variables and time-series features to accurately predict future prices and trends. This approach incorporates both the inherent temporal patterns of prices and the causal influence of exogenous variables, improving forecast accuracy, providing a reliable price basis for coal blending and procurement decisions, and reducing cost risks caused by price fluctuations.

[0011] Furthermore, the step of generating a procurement plan suggestion corresponding to the coal blending scheme based on the price trend of raw coal and the inventory information includes: The price fluctuation coefficient is determined based on the stated price trend of raw coal; Extract the proportion, quality fluctuation coefficient, and planned consumption of each type of raw coal in the candidate coal blending scheme; Calculate the lower limit of safe inventory and the upper limit of inventory warning for raw coal. The lower limit of safe inventory = average daily consumption × (procurement cycle + buffer days) × (1 + quality fluctuation coefficient) × (1 + price fluctuation coefficient). The upper limit of inventory warning = lower limit of safe inventory × (1 + price fluctuation buffer ratio). When the inventory information is less than the preset percentage of the safety stock lower limit, a shortage warning message is generated, and the purchase quantity is determined as (planned consumption - inventory) × (1 + emergency buffer ratio). When the inventory information is between the lower limit of the safety stock and the upper limit of the inventory warning, an early warning message is generated, and the purchase quantity is determined as (purchase cycle × average daily consumption) × fluctuation coefficient. When the inventory information is greater than or equal to the inventory warning limit, a safety alert is generated, and the optimal purchasing period is determined based on the price trend.

[0012] By adopting the above technical solutions, a quality fluctuation coefficient is introduced into the safety stock lower limit formula to cope with the additional consumption caused by unstable coal quality. The upper limit of the inventory warning is set with a price fluctuation buffer ratio to prevent the risk of price increases in advance. The emergency buffer ratio reserves procurement flexibility for emergencies, enhances the resilience of the supply chain, and the tiered response strategy ensures production continuity, maintains a reasonable inventory level, and reduces costs.

[0013] Furthermore, if the coal blending scheme includes multiple schemes, the method further includes: The coke quality indicators for each coal blending scheme are determined using an expert-guided model. Calculate the predicted cost price corresponding to the coal blending scheme based on the predicted price of each type of raw coal; The quality indicators and predicted cost prices of each coal blending scheme are dynamically weighted to calculate the total evaluation score, and the evaluation priority sequence is generated by sorting the total evaluation scores in descending order.

[0014] By employing the above technical solution, this method determines quality indicators for multiple coal blending schemes through expert models, calculates costs by combining predicted prices, and then calculates and ranks the total evaluation score according to dynamic weights. This achieves the quantification of the merits of the schemes, facilitating the rapid selection of a scheme that balances quality and cost.

[0015] Furthermore, before calculating the total evaluation score by assigning dynamic weights to the quality indicators and predicted cost prices of each coal blending scheme, the method also includes: The causal influence of exogenous variables on raw coal prices is normalized, and a static causal weight vector is output. The exogenous variables and historical price data are spliced ​​together in time sequence to form the time series data of the exogenous variables; Construct an LSTM dynamic weight allocation model, add a fully connected layer before the LSTM dynamic weight allocation model, and use the static causal weight vector as the learnable initial weights. The LSTM dynamic weight allocation model is applied, with exogenous variable time series data as input, and dynamic weight coefficients as output. The dynamic weights are obtained by multiplying the static causal weight vector and the dynamic weight coefficients element by element.

[0016] By adopting the above technical solution, a static weight vector is generated by normalizing the causal influence intensity, which transforms the influence of exogenous variables on prices into quantifiable weight parameters, enhancing the interpretability of the model. The LSTM dynamic weight allocation model can learn the implicit patterns in the time series data of exogenous variables and dynamically adjust the weights according to the market state. By multiplying the static causal weights with the dynamic weight coefficients, the basic influence of the causal relationship of variables is preserved, while incorporating the dynamic characteristics of time series changes, achieving a balance between stability and flexibility.

[0017] Furthermore, the optimization suggestions include suggestions for changing the coal type. The expert-guided model determines the optimization suggestions based on the coal blending scheme, including: The expert-guided model is used to analyze the coal quality test data of the raw coal to determine the quality index of the raw coal; The effects of the characteristics of each type of raw coal on coke quality were analyzed and determined. Raw coal with a quality index below the index threshold and / or that would reduce coke quality was screened as problematic raw coal. Obtain alternative raw coal that can replace the problematic raw coal, wherein the quality index of the alternative raw coal is not lower than the index threshold and / or improves the quality of coke; The problematic raw coal is replaced by the alternative raw coal to obtain a variety of new raw coals for blending. An additional blending scheme is generated based on the new raw coals for blending. A suggestion for changing coal types is generated based on the alternative raw coals and the additional blending scheme.

[0018] By adopting the above technical solutions and analyzing coal quality through expert models, problematic coal types affecting coke quality are identified, high-quality replacement coals are matched, and new solutions are generated, resulting in coal replacement recommendations. This can improve coke quality, optimize coal blending schemes, and ensure the effectiveness of coal blending.

[0019] Furthermore, the method also includes: Obtain historical price data for each type of raw coal; Calculate the coal price volatility based on the historical price data; Based on the option pricing model, the coal price volatility is converted into an option premium; Based on the historical price data and the coal price volatility, a probability distribution path for coal prices is generated through Monte Carlo simulation. Based on the aforementioned coal price distribution path and option premium, calculate the optimal hedging ratio between futures and spot prices; Perform dynamic hedging operations based on the optimal hedging ratio.

[0020] By adopting the above technical solution, coal price volatility is calculated using historical price data to capture market uncertainty. Monte Carlo simulation generates probability distribution paths, fully considering the randomness and path dependence of coal price fluctuations. This breaks through the limitations of traditional static models. By combining price paths and option premiums to calculate the optimal hedging ratio, a balance is achieved between minimizing risk exposure and optimizing hedging costs.

[0021] Furthermore, the real-time acquisition of coal quality testing data, market prices, and inventory information for each type of raw coal includes: Real-time coal quality testing data is obtained through the production management system; Use web crawling technology to capture real-time market prices; The system acquires dynamic scanning data of coal piles in the raw coal warehouse, combines it with computer vision algorithms to identify the volume of coal piles, and determines the first inventory data in the raw coal warehouse. Obtain the location and coal transport volume of vehicles awaiting unloading and vehicles en route; Based on the vehicle's location, the time when new raw coal is delivered to the warehouse is predicted; Based on the raw coal's arrival time and the amount of coal transported on the vehicles, real-time warehousing data is calculated. This real-time warehousing data is then added to the first inventory data to obtain real-time inventory information. Predicting the new arrival time of raw coal based on the vehicle's location includes: Get vehicle speed; Determine the distance between the vehicle's location and the unloading endpoint; The transportation time is calculated based on the vehicle speed and the distance, and the raw coal storage time is calculated based on the transportation time and the floating time. If each historical delivery time is less than the average delivery time, then the fluctuation time is zero. If the proportion of historical delivery time exceeding the average delivery time reaches a first preset value, then calculate the difference between each historical delivery time and the average delivery time, calculate the mean of each difference to obtain the mean difference, obtain the vehicle's delivery delay rate within the preset time period, and calculate the floating time = mean difference × (1 + delivery delay rate / 100%).

[0022] By adopting the above technical solutions, through direct connection to the production management system, crawling technology to capture and dynamically scan data, the coal quality, price and inventory data can be updated in minutes or even seconds, eliminating the lag of traditional manual data entry. Combined with data on vehicles in transit, a full-chain visualization of "current inventory + in-transit inventory" is formed. Based on location data, the warehouse entry time can be predicted, and warehouse space and unloading resources can be planned in advance.

[0023] In summary, this application includes at least one of the following beneficial technical effects: 1. A coal blending decision-making system is constructed by integrating various information such as raw coal. With expert guidance and mathematical models, the system generates and optimizes the blending scheme, ensuring quality and cost control. It also generates procurement recommendations based on market forecasts and produces reports. This intelligent coal blending system, integrating a mathematical model, an expert-guided model, and a market forecasting model, overcomes the limitations of traditional reliance on manual experience. It achieves fully automated decision-making from data input to optimal formula generation, enabling scientific optimization of the scheme, reducing supply chain risks, and improving intelligence and economy. 2. The mathematical calculation model and the expert-guided model have undergone multiple interactions and adjustments, considering objective factors such as the chemical composition and physical properties of the raw coal, while also fully incorporating the experience and knowledge of experts, making the blending results more scientific and reasonable. 3. By applying market forecasting models to capture market dynamics and predict future market prices, and by providing suggestions on raw coal procurement plans based on changes in market conditions, enterprises can effectively avoid market risks and optimize inventory management while ensuring production needs are met.

[0024] 4. Achieve integrated management driven by data from raw material procurement to coke production. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating the intelligent prediction method for coking coal feed in the embodiments of this application.

[0026] Figure 2 This is a schematic diagram of the structure of the intelligent prediction system for coal feeding in coking plants, as described in this application. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0028] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article, unless otherwise specified, generally indicates that the preceding and following related objects have an "or" relationship.

[0029] This application discloses an intelligent prediction method for coal feed into coking plants. (Refer to...) Figure 1 This is performed by an electronic device, which can be a server or a terminal device. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The terminal device can be a smartphone, tablet, desktop computer, etc., but is not limited to these. This includes (steps S101 to S107): Step S101: Obtain the quality and cost indicators of various raw coals and cokes used for coal blending.

[0030] Specifically, the electronic equipment obtains information about various raw coals for coal blending based on the operator's input; that is, the operator inputs or selects the raw coals to be used in the blending process through the operating interface. The operating interface can pre-save data on various raw coals for easy selection by the operator.

[0031] The quality and cost indicators for coke are determined based on the company's production needs and product quality requirements. Coke quality indicators mainly include ash content, sulfur content, cold strength, hot strength, and physical properties such as particle size and caking properties. When setting these indicators, the electronic device provides a corresponding interface and operation prompts to facilitate user settings according to actual conditions. Once set, the indicators will serve as the basis for subsequent coal blending calculations and model evaluations.

[0032] Step S102: Obtain real-time coal quality testing data, market price, and inventory information for each type of raw coal.

[0033] Further, step S102 includes steps S1021 to S1026): Step S1021: Obtain coal quality testing data for each type of raw coal in real time through the production management system.

[0034] Specifically, coal quality testing data includes multiple key parameters such as coal type, ash content, sulfur content, volatile matter, caking index, moisture content, and fineness, comprehensively reflecting the quality and characteristics of the raw coal and serving as the foundational data for coal blending calculations. Staff obtain coal quality testing data in the laboratory based on the current raw coal testing, and electronic equipment acquires this data through a data acquisition system.

[0035] Step S1022: Use web crawling technology to capture real-time market prices.

[0036] Step S1023: Obtain dynamic scanning data of coal piles in the raw coal warehouse, identify the volume of coal piles using computer vision algorithms, and determine the first inventory data in the raw coal warehouse.

[0037] Specifically, inventory information includes the current inventory quantity, location, and arrival time of each type of raw coal. The accuracy of inventory data is crucial for subsequent coal blending calculations and procurement recommendations. It helps the system understand the supply of raw coal in real time, avoiding impacts on production and costs due to insufficient or excessive inventory.

[0038] Inventory information allows electronic devices to monitor the real-time inventory levels of various raw coals, ensuring that inventory factors are fully considered during coal blending and preventing insufficient or excessive inventory of any particular type of coal from affecting the blending plan. To facilitate dynamic acquisition of inventory data, the electronic devices obtain partial inventory data from the raw coal warehouses, further considering inventory that has not yet been stored during transportation.

[0039] To obtain warehouse inventory, point cloud scanning equipment is installed in the raw coal warehouse. The point cloud scanning equipment scans the coal pile in real time, and the scanned data is dynamic scan data. Combined with computer vision algorithms, the volume of the coal pile is obtained based on the dynamic scan data and used as the first inventory data.

[0040] Step S1024: Obtain the vehicle location and coal transport volume of the vehicles to be unloaded and the vehicles in transit.

[0041] Specifically, in order to obtain transportation inventory, each vehicle shares its location and coal transport volume with the manufacturer via the cloud, thereby enabling electronic devices to obtain relevant information about vehicles waiting to be unloaded.

[0042] Step S1025: Based on vehicle location, predict the time when raw coal will be delivered to the warehouse.

[0043] Specifically, the electronic equipment determines the distance between the vehicle's location and the unloading endpoint based on the vehicle's position, calculates the transportation time based on the vehicle's speed, and calculates the raw coal storage time based on the transportation time and the floating time.

[0044] In calculating the floating time, the historical delivery time of the current vehicle is compared with the average delivery time. If all historical delivery times are less than the average delivery time, the floating time is zero. If the proportion of historical delivery times that are greater than the average delivery time reaches a first preset value, the difference between each historical delivery time and the average delivery time is calculated, and the mean of each difference is calculated to obtain the average difference. The delivery delay rate of the vehicle within the preset time period is obtained, and the floating time is calculated as: average difference × (1 + delivery delay rate / 100%).

[0045] Step S1026: Calculate the real-time warehousing data based on the raw coal warehousing time and the amount of coal transported on the warehousing vehicle. Add the real-time warehousing data to the first inventory data to calculate the real-time inventory information.

[0046] Furthermore, during the data import process, the system performs format checks and integrity verification to ensure that the imported data meets the system's requirements. For data that does not meet the requirements, the system will prompt the user to correct or supplement it to ensure data quality and reliability.

[0047] For example, sensor malfunctions may lead to outliers in coal quality testing data; sudden market fluctuations may cause price deviations, resulting in abnormal fluctuations in market price data; and abnormal stockout records during seasonal fluctuations may lead to abnormal data in inventory information. These abnormal data may reduce the accuracy of coking coal input predictions, so it is necessary to remove data noise to provide a high-quality data foundation for subsequent calculations.

[0048] Therefore, after acquiring data, electronic devices use a hybrid algorithm based on Bayesian probability models and robust outlier detection to remove data noise from coal quality testing data, market prices, and inventory information.

[0049] The data undergoes Bayesian probability modeling and robust statistical preprocessing. The outputs of both are then combined, followed by posterior probability anomaly assessment to remove noise and generate clean data, including (steps S11–S16): Step S11: Standardize the data to eliminate the influence of dimensions.

[0050] Step S12: Set a robust threshold, initially identify outliers that deviate from the robust threshold, and mark the outlier data points. After the data points have passed the robust outlier monitoring, if they are outliers, the robust normalization value is 1; otherwise, the robust normalization value is 0.

[0051] Step S13: Construct a Bayesian probability model and set prior probability distributions for each indicator based on historical data. That is, select the corresponding prior distribution based on the characteristics of coal quality testing data, market prices, and inventory data, etc. For example, if the distribution characteristics of coal quality testing data are approximately normal, then the Bayesian prior will choose the normal distribution. Where μ0 is the historical mean and σ0 is the standard deviation; the market price distribution is right-skewed and has a peak, so the Bayesian prior is a gamma distribution; the inventory information distribution is cyclical, so a seasonal Bayesian model is chosen.

[0052] Step S14: Set the likelihood function, determine the anomalous prior probability of the data points following the corresponding prior distribution, further determine the probability density of the outlier data, and calculate the posterior anomalous probability of the data points using Bayes' theorem based on the anomalous prior probability and probability density.

[0053] Step S15: Define a hybrid outlier score, assigning weights to the data points obtained through Bayesian formula and robust outlier monitoring, for example, outlier score S(x) = ω a · Robust standardized value + ω b • Posterior anomaly probability, where ω a and ω b The weights sum to 1.

[0054] Step S16: When the outlier score S(x) > the mixing threshold, the data point is determined to be a noise point and the data point is removed.

[0055] Step S103: Apply the expert-guided model to conduct in-depth analysis of the characteristics of raw coal based on coal quality testing data, determine the analysis results of raw coal, and determine the constraints based on the quality indicators and cost indicators of raw coal and coke.

[0056] Specifically, the electronic equipment is pre-trained with an expert guidance model. The expert guidance model in this application can simulate the knowledge and experience of experts, conduct in-depth analysis of raw coal, evaluate the rationality of coal blending schemes, and provide decision support during the coal blending process.

[0057] To enable the expert-guided model to perform the aforementioned functions, the electronic equipment first acquires a large amount of training data, including historical formula data and expert evaluation data. The historical formula data comes from the company's long-term accumulated historical coal blending data, including raw coal quality testing data under different production conditions, raw coal ratios, corresponding coke quality indicators, and information on anomalies during the production process. This data reflects the company's past production experience and decision-making processes. The expert evaluation data includes evaluations and scores of historical formulas by senior coking experts, as well as information on raw coal ratios under production conditions, corresponding coke quality indicators, and anomalies during the production process.

[0058] The electronic device divides historical formula data and expert evaluation data into training and testing sets, inputs them into a machine learning model for training and fitting, verifies and tunes parameters, and then uses the testing set to evaluate accuracy, ultimately obtaining a trained expert evaluation model. The machine learning model can be a decision tree, random forest, or Bayesian model, among others.

[0059] Therefore, after the electronic equipment inputs the raw coal into the expert-guided model, the model performs an in-depth analysis of the raw coal, judging the quality, applicability, and potential problems of each type of coal, and obtaining the raw coal analysis results. For example, it analyzes the impact of excessively high or low caking index of a certain type of coal on coal blending schemes and coke quality, providing a reference for coal blending decisions.

[0060] Furthermore, the expert-guided model provides constraints, enabling the mathematical calculation model to be performed under these constraints during coal blending. This results in a scientifically accurate coal blending scheme that better meets the requirements for coke quality, cost, and environmental protection.

[0061] Step S104: Based on the coal quality testing data, market price, and inventory information of each type of raw coal, a preliminary coal blending scheme is generated using a mathematical calculation model employing mixed integer rules and a genetic algorithm, with the goal of minimizing coke quality indicators and costs, and in conjunction with constraints. This includes (steps S1041 to S1045): Step S1041: Construct the objective function using mixed-integer programming; The objective function is: minF=ω1C+ω2Q(1) C=∑ i,t x ijt ·p i ·m t (2) Where C is the coal blending cost, x ijt Let p be the proportion of the i-th type of raw coal in the j-th coal blending scheme during the t-th period. i m is the unit price of the i-th type of raw coal. tWhere Q represents the total amount of coal to be blended, and Q is the penalty for coke quality deviation. Let k be the measured value of the k-th quality indicator. Let λ be the target value of the k-th quality indicator. k ω1 and ω2 are the weights corresponding to the k-th quality indicator, respectively, and the weight coefficients for coal blending cost and coke quality.

[0062] The constraints are: I it =I i,t-1 +y it ·s it -x ijt ·m t ≥I min (5) Among them, a ik Let b be the k-th index value of the i-th type of raw coal. k I represents the upper or lower limit of the value of the k-th indicator. it Let s be the inventory of the i-th type of raw coal in period t. it Let y be the purchase quantity of the i-th type of raw coal in period t. it I is an indicator variable for whether the i-th type of raw coal was purchased. min This is for safety stock levels.

[0063] Specifically, objective function (1) sets weight coefficients ω1 and ω2 according to actual needs, representing the combined goals of achieving coke quality standards and minimizing costs; objective function (2) represents the product of the total amount of the i-th raw material in the j-th coal blending scheme and the unit price of the i-th raw material, and then calculates the coal blending cost; objective function (3) represents the calculation of quality deviation when the measured value of the k-th quality indicator is greater than the target value.

[0064] The constraints are the rules that need to be followed in achieving the objective function. (4) In the raw coal, the sum of the k-th index value of the i-th raw coal in the j-th coal blending scheme cannot exceed the upper or lower limit of the corresponding index value; (5) The inventory of the i-th raw coal in the t-th period is the sum of the inventory and purchase amount of the previous period minus the total amount of coal blending; (6) In a single coal blending scheme, the sum of the proportions of various raw coals is 1.

[0065] Step S1042: Based on inventory information and coal quality testing data, randomly generate N feasible coal blending schemes that satisfy the objective function and constraints.

[0066] Step S1043: For each feasible coal blending scheme, the ratio and type of raw coal of the feasible coal blending scheme are iteratively optimized by genetic algorithm to meet the constraints.

[0067] Step S1044: After each iteration, calculate the fitness of the feasible coal blending scheme based on market price and inventory information, combined with the objective function; Step S1045: After multiple rounds of iterative optimization, at least one feasible coal blending scheme with a fitness value greater than the preset value is determined as the coal blending scheme.

[0068] Specifically, the proportions and types of raw coal for each feasible coal blending scheme are encoded. For example, real number encoding can be used, where the proportion of each type of raw coal is represented by a real number, and different combination sequences of different coal types correspond to different encoding sequences. N randomly generated feasible coal blending schemes are used as the initial population, with each feasible coal blending scheme being an individual within the population.

[0069] A fitness function is designed based on the objective function to evaluate the merits of each coal blending scheme. The objective function value can be modified to obtain the fitness value. For example, if the objective function is to minimize the cost, the fitness function can be set as F = 1 / (objective function value + ε). To prevent the denominator from being 0, ε is a local minimum value. Therefore, the smaller the objective function value, the higher the fitness.

[0070] Based on individual fitness, methods such as roulette wheel selection and tournament selection are used to select individuals with high fitness from the current population to enter the next generation. Individuals with higher fitness have a greater probability of being selected. Crossover operations are then performed on the selected individuals to simulate the genetic recombination process in organisms.

[0071] Mutation operations simulate gene mutations in biological evolution by mutating an individual's genes with a certain mutation probability. For example, a gene of an individual can be randomly selected and slightly altered, such as changing 0.3 in [0.2,0.3,0.1,0.25,0.15] to 0.32. Mutation operations can increase the diversity of the population and prevent the algorithm from getting trapped in local optima too early.

[0072] Repeated selection, crossover, and mutation operations generate a new population, iterating continuously until a stopping condition is met, such as reaching a preset maximum number of iterations, or the fitness values ​​of individuals in the population showing little change and tending towards stability. After the iteration ends, individuals with fitness values ​​greater than a preset value are selected from the last generation of the population; their corresponding coal blending scheme is the optimal or near-optimal coal blending scheme optimized by the genetic algorithm.

[0073] Step S105: Based on multiple interactions between the mathematical calculation model and the expert guidance model, adjust the coal blending scheme. The expert guidance model determines optimization suggestions based on the coal blending scheme.

[0074] Specifically, after the coal blending mathematical model completes its initial calculations, it obtains the reasonable range for the proportion of each type of raw coal. At this point, experts guide the model to begin analyzing these results.

[0075] In addition to analyzing raw coal, the expert-guided model can also score the rationality of the formulas calculated by the coal blending mathematical model. The scoring criteria include multiple dimensions such as whether the formula meets the target parameter requirements, whether the raw coal blending is reasonable, whether production costs are controllable, and whether it complies with environmental protection requirements. Formulas with scores below a threshold are then removed, eliminating those that theoretically meet the requirements but may have problems in actual production.

[0076] After screening coal blending schemes, the expert-guided model identifies coal types with unreasonable proportions or poor quality based on historical experience and current data, and then provides reasonable adjustment suggestions, such as fine-tuning the proportion of a certain coal or changing the type of coal.

[0077] For example, if the sulfur content of a certain type of coal exceeds the standard and the cost is too high, the modification suggestion is to reduce the proportion of coal type A to below 15%; or, if the coal blending scheme can theoretically achieve the coke index, but the actual coke effect cannot achieve the theoretical result, it is recommended to change the coal type or adjust the proportion of a certain type of coal.

[0078] For changing the coal type, the process includes (steps S21 to S25): Step S21: Apply expert-guided model analysis to determine the quality index of raw coal by analyzing the coal quality test data.

[0079] Specifically, experts guide the model to pre-set a scoring formula, determine the score of raw coal based on coal quality testing data, and the resulting score is the quality index.

[0080] Step S22: Analyze and determine the impact of the characteristics of each raw coal on coke quality.

[0081] Specifically, the expert-guided model uses experience to determine the impact of each type of raw coal on coke quality, such as improving or reducing coke quality.

[0082] Step S23: Select raw coal with a quality index below the index threshold and / or that would reduce the quality of coke as problematic raw coal.

[0083] Step S24: Obtain a substitute raw coal that can replace the problematic raw coal, wherein the quality index of the substitute raw coal is not lower than the index threshold and / or will improve the quality of coke.

[0084] Specifically, the expert-guided model selects suitable replacement raw coal based on experience, and can also use coal types that have received positive feedback in the market as replacement raw coal.

[0085] Step S25: Replace the problematic raw coal with the alternative raw coal to obtain a variety of new raw coals for blending. Generate an additional blending scheme based on the new raw coals for blending. Generate a coal type replacement suggestion based on the alternative raw coals and the additional blending scheme.

[0086] During the interaction between the coal blending mathematical model and the expert-guided model, the two continuously exchange data and share information. The coal blending mathematical model will revise the reasonable range of proportions based on the analysis results of the expert-guided model; while the expert-guided model will also refer to the calculation data of the coal blending mathematical model to further optimize its analysis results.

[0087] After multiple rounds of interaction and adjustments, the optimal blending ratio for each type of raw coal was finally determined. This ratio took into account both objective factors such as the chemical composition and physical properties of the raw coal, and fully incorporated the experience and knowledge of experts, resulting in a more scientific and reasonable blending outcome.

[0088] Furthermore, the coal blending scheme can have multiple formulas that meet both theoretical and empirical standards. The electronic equipment can select the most effective formula, that is, the coal blending scheme with the lowest cost and the best stability while meeting quality and production requirements.

[0089] Step S106: Apply a market forecasting model based on LSTM neural network and time-series causal analysis to predict the price trend of raw coal in the future within a preset time period. Based on the price trend and inventory information of raw coal, generate procurement plan suggestions corresponding to the coal blending plan.

[0090] Specifically, the market forecasting model is an intelligent analysis tool that predicts the price trend of coking coal based on the final coal blending scheme. It combines market supply and demand data, historical price information, macroeconomic indicators and other factors to predict future market prices.

[0091] To obtain a more accurate market forecasting model, the model needs to be trained with a dataset consisting of factors affecting raw coal prices. In order to screen these factors, time-series causal analysis is used to determine Granger causal variables, and then the relevant information of the current Granger causal variables is obtained to train the market forecasting model, as in steps S31 to S36, and then the price trend is obtained.

[0092] Step S31: Apply causal tests to quantify the causal influence of multiple exogenous variables on raw coal prices.

[0093] Specifically, the explained variable is the price of raw coal, and the exogenous variables include relevant information about the domestic and international coking coal markets, including coal price trends, capacity changes, import and export data, transportation costs, macroeconomic indicators (such as GDP growth, steel industry demand, etc.), policies and regulations, etc.

[0094] If historical information about a variable can significantly improve the accuracy of predicting raw coal prices, then the Granger causality test should be applied to determine the strength of the causal influence of that exogenous variable on raw coal prices. For example, a causal influence strength of 0.5 for crude oil prices means that "for every 1 standard deviation fluctuation in crude oil prices, the average fluctuation in raw coal prices is 0.5 standard deviations."

[0095] Step S32: Obtain historical price data for each type of raw coal.

[0096] Step S33: Obtain the time series features of historical price data and the causal influence strength of the corresponding exogenous variables, and standardize the time series features to obtain a dataset consisting of historical price data, its corresponding time series features, and Granger causal variable data.

[0097] Specifically, time series characteristics include static features such as mean, maximum, minimum, standard deviation, and variance, as well as trend features, periodic features, volatility characteristics, and abrupt change features.

[0098] Because the dimensions and value ranges of different time series features vary considerably, Z-score standardization is used to transform the feature values. μ is the mean of the feature, and σ is the standard deviation.

[0099] Standardization preserves the original data distribution and ensures that different features have the same weight during model training, thus improving the model's prediction accuracy.

[0100] Finally, a dataset was constructed based on historical price data, time series characteristics, and Granger causal variable data.

[0101] Step S34: Construct an LSTM model, use the dataset to train the LSTM model, and obtain the trained LSTM market price prediction model.

[0102] Specifically, the LSTM model training process begins with initializing the parameters, iteratively inputting the dataset, calculating the predicted value through forward propagation, tuning the parameters through backpropagation, and optimizing through multiple rounds of iteration until convergence, thus obtaining the trained LSTM model for market price prediction.

[0103] Step S35: Obtain real-time Granger causal variable data and time series features corresponding to market prices.

[0104] Step S36: Input the market price of raw coal, the time series characteristics corresponding to the market price, and the real-time Granger causal variable data into the trained LSTM market price prediction model to predict the predicted prices at multiple future moments within a preset future time period, and obtain the price trend based on the predicted prices at each future moment.

[0105] Specifically, the LSTM market price forecasting model can consider short-term (e.g., weekly, monthly) and long-term (e.g., quarterly, annual) price trends, providing forward-looking information for companies' purchasing decisions. For example, by analyzing factors such as seasonal demand changes and capacity fluctuations in major coal-producing areas, it can predict the upward or downward trend of coking coal prices in the future.

[0106] By combining market dynamics and actual business needs, the price forecast results are continuously revised and optimized. As market conditions change and new data is added, the model can adjust the forecast results in a timely manner, improving the accuracy and reliability of the forecasts.

[0107] Furthermore, market forecasting models can provide enterprises with procurement recommendations for each type of raw coal based on price trends and inventory information, taking into account factors such as the enterprise's production needs, financial situation, and storage capacity, and formulate personalized procurement plans.

[0108] Therefore, it is essential to develop a scientific inventory plan that balances inventory costs with production assurance. Based on price forecasts and production plans, determine the safe inventory levels for various raw coals to avoid production interruptions due to insufficient inventory or excessive capital tied up in unsold inventory. The method includes (steps S41–S46): S41: Determine the price fluctuation coefficient based on the price trend of raw coal.

[0109] Specifically, the electronic device calculates the price volatility based on the price trend of raw coal within a time period. The volatility is calculated as (highest price in the period - lowest price in the period) / average price in the period, and the price volatility is determined as the price volatility coefficient.

[0110] Step S42: Extract the proportion, quality fluctuation coefficient, and planned consumption of each type of raw coal in the candidate coal blending scheme.

[0111] Specifically, the planned consumption of each type of raw coal is the product of the total amount of coal to be blended in the candidate coal blending scheme and the proportion of each type of raw coal.

[0112] Step S43: Calculate the lower limit of safe inventory and the upper limit of inventory warning for raw coal. Lower limit of safe inventory = average daily consumption × (procurement cycle + buffer days) × (1 + quality fluctuation coefficient) × (1 + price fluctuation coefficient). Upper limit of inventory warning = lower limit of safe inventory × (1 + price fluctuation buffer ratio).

[0113] Specifically, for coal types with large price fluctuations and relatively tight supply, safety stock should be appropriately increased; for coal types with stable prices and sufficient supply, a low inventory level should be maintained.

[0114] Step S44: When the inventory information is less than the preset percentage of the safety stock lower limit, a shortage warning message is generated, and the purchase quantity is determined as (planned consumption - inventory) × (1 + emergency buffer ratio).

[0115] Step S45: When the inventory information is between the lower limit of safety stock and the upper limit of inventory warning, generate a warning message and determine the purchase quantity = (purchase cycle × average daily consumption) × fluctuation coefficient.

[0116] Step S46: When the inventory information is greater than or equal to the inventory warning limit, a safety alert message is generated, and the best time to purchase is determined by the price trend.

[0117] Specifically, the inventory and procurement plan generation module generates different warnings or prompts based on the relationship between inventory information and the lower limit and upper limit of safety stock, making it easier for staff to adjust procurement based on the current inventory.

[0118] When price trends are predicted using an LSTM neural network, it is recommended to increase the purchase volume appropriately during price troughs to reduce procurement costs; during price peaks, the purchase pace should be reasonably controlled to avoid excessively high procurement costs.

[0119] Furthermore, the procurement plan recommendations also include cost-effective coal type recommendations. The market forecasting model compares the price, quality, and contribution to coke quality of different coal types to recommend more cost-effective alternatives on the market. When proposing replacement recommendations, the model will comprehensively consider factors such as the chemical composition, physical properties, and coking properties of the replacement coal type to ensure that the formula after replacement can still meet the coke quality indicators and production requirements.

[0120] For example, if the price of a certain type of coal rises sharply in the market, while another type of coal, similar in chemical composition and coking properties but cheaper, is available, the market forecasting model will suggest appropriately increasing the proportion of the lower-priced coal in the blending formula to reduce blending costs. Simultaneously, the model will recalculate and evaluate the formula based on actual conditions to ensure the rationality and feasibility of the replacement formula. This helps companies optimize their formulas, reduce production costs, and ensure that coke quality is not affected.

[0121] Furthermore, if there are multiple coal blending schemes, the electronic equipment can combine the results from expert guidance models and market forecasting models to prioritize the various coal blending schemes, making it easier for staff to select the appropriate one. This includes (steps S51 to S53): Step S51: Apply expert-guided models to determine the coke quality indicators for each coal blending scheme.

[0122] Step S52: Calculate the corresponding predicted cost price based on the predicted price of each type of raw coal.

[0123] Specifically, the predicted price of the raw coal for the i-th type of coal is taken as p. i Substituting this into the objective function (2) above, we obtain the predicted cost price C.

[0124] Step S53: Calculate the total evaluation score by assigning dynamic weights to the quality indicators and predicted cost prices of each coal blending scheme, and generate an evaluation priority sequence by sorting the total evaluation scores in descending order.

[0125] Specifically, before calculating the total evaluation score, steps (Sa to Se) are performed: Step Sa: Normalize the causal influence of exogenous variables on raw coal prices and output a static causal weight vector.

[0126] Step Sb: Synchronize and concatenate the exogenous variables with historical price data over time to form time-series data of the exogenous variables.

[0127] Specifically, the static causal weight vector reflects the long-term average influence of exogenous variables. Exogenous variables are aligned with historical data on raw coal prices according to timestamps to form a multivariate time series.

[0128] Step Sc: Construct an LSTM dynamic weight allocation model, add a fully connected layer before the LSTM dynamic weight allocation model, and use the static causal weight vector as the learnable initial weights.

[0129] Step Sd: Apply the LSTM dynamic weight allocation model, input the time series data of exogenous variables, and output the dynamic weight coefficients.

[0130] Specifically, the fully connected layer takes static causal weights as input and outputs the initial weights of the science system. The LSTM layer receives exogenous variable time-series data, captures the dynamic features of the time series, and then maps the LSTM hidden states to dynamic weight coefficients through the fully connected layer.

[0131] The exogenous variable time series data is input into the LSTM to obtain the hidden state at each time step, and the dynamic weight coefficients are calculated through the output layer.

[0132] Step Se: Multiply the static causal weight vector and the dynamic weight coefficients element by element to obtain the dynamic weights.

[0133] Specifically, when the dynamic weight coefficient is large, the influence of the corresponding exogenous variable is strengthened, such as a sudden policy causing a surge in the weight of a certain variable. When the dynamic weight coefficient is small, the influence is weakened, such as seasonal factors being suppressed during irrelevant periods.

[0134] Dynamic weights combine static weights with time-series dynamic modeling, avoiding the causal confusion problem of pure data-driven models. They can dynamically adjust the influence weights of exogenous variables according to changes in the market environment, thus improving forecasting flexibility.

[0135] Step S107: Generate a coal intake evaluation report that includes at least the raw coal analysis results, coal blending scheme, scheme optimization suggestions, and procurement scheme suggestions.

[0136] Specifically, the electronic equipment integrates the above analysis results and outputs a coal intake evaluation report to provide decision-making reference for staff.

[0137] Furthermore, the electronic equipment is equipped with a procurement option hedging model to generate the optimal hedging ratio between futures and spot prices, ensuring production supply and improving economic efficiency. The method includes (steps S61 to S66): Step S61: Obtain historical price data for each type of raw coal.

[0138] Specifically, the electronic device interacts with multiple data sources, including but not limited to coal trading platforms, industry databases, and financial data providers, to collect price information on raw coal of different times, regions, and qualities. To ensure the accuracy and completeness of the data, the module cleans and verifies the collected data. For example, it removes duplicate data, corrects erroneous data, and fills in missing data.

[0139] Step S62: Calculate the coal price volatility based on historical price data.

[0140] Specifically, electronic devices measure the degree of price volatility by calculating the standard deviation or variance of prices.

[0141] Step S63: Based on the option pricing model, convert coal price volatility into option premium.

[0142] Specifically, common option pricing models include the Black-Scholes model and its extensions. These models use coal price volatility as one of the input parameters, combined with other relevant parameters such as the option's strike price, time to expiration, and risk-free interest rate. Through complex mathematical calculations, the theoretical value of the option, i.e., the option premium, is derived. The higher the volatility, the higher the option premium typically is.

[0143] Step S64: Generate the probability distribution path of coal prices through Monte Carlo simulation based on historical price data and coal price volatility.

[0144] Specifically, based on historical price data and coal price volatility, a probability distribution model for coal prices is determined, such as the normal distribution or log-normal distribution. A large number of coal price sample paths are generated through random sampling to simulate possible future changes in coal prices. These sample paths constitute the probability distribution path of coal prices, which can intuitively show the possible range and probability distribution of coal prices at different points in the future.

[0145] Step S65: Combine the coal price distribution path and option premium to calculate the optimal hedging ratio between futures and spot.

[0146] Specifically, considering various factors such as coal price fluctuations, option costs and returns, and the price relationship between futures and spot markets, mathematical models and optimization algorithms are used to simulate and evaluate different hedging ratios to find the optimal hedging ratio that achieves the best balance between risk and return. The mathematical model may use risk metrics such as Value at Risk (VaR) and Conditional Value at Risk (CVaR) to assess the risk level under different hedging ratios.

[0147] Step S66: Perform dynamic hedging operations according to the optimal hedging ratio.

[0148] Specifically, by integrating with the trading system, it automatically executes buying and selling operations for futures and spot commodities, and adjusts the ratio of futures to spot commodities in a timely manner according to changes in market conditions to maintain the optimal hedging ratio and achieve effective risk management.

[0149] This application also provides an intelligent prediction system for coking coal feed, referring to... Figure 2 The intelligent prediction system for coal feeding in coking plants includes: The basic data acquisition module 201 is used to acquire various raw coal and coke quality and cost indicators for coal blending; The real-time data acquisition module 202 is used to acquire coal quality test data, market price and inventory information for each type of raw coal in real time; the constraint determination module 203 is used to apply an expert-guided model to conduct in-depth analysis of the characteristics of raw coal, determine the analysis results of raw coal, and determine the constraint conditions based on the quality indicators and cost indicators of raw coal and coke. The preliminary coal blending scheme generation module 204 is used to generate a preliminary coal blending scheme based on the coal quality test data, market price and inventory information of each raw coal, using a mathematical calculation model with mixed integer rules and genetic algorithms, with the goal of minimizing coke quality indicators and costs, combined with constraints. The coal blending scheme optimization module 205 is used to adjust the coal blending scheme through multiple interactions between the mathematical calculation model and the expert guidance model. The expert guidance model determines the scheme optimization suggestions based on the coal blending scheme. The market forecasting model application module 206 is used to apply a market forecasting model based on LSTM neural network and time series causal analysis to predict the price trend of raw coal in the future within a preset time period, and generate procurement plan suggestions corresponding to the coal blending plan based on the price trend and inventory information of raw coal. The coal intake evaluation report generation module 207 is used to generate a coal intake evaluation report that includes at least the raw coal analysis results, coal blending plan, plan optimization suggestions, and procurement plan suggestions.

[0150] Furthermore, the preliminary coal blending scheme generation module 204 is specifically used for: Construct the objective function using mixed-integer programming; The objective function is: minF=ω1C+ω2Q C=∑ i,t x ijt ·p i ·m t Where C is the coal blending cost, x ijt Let p be the proportion of the i-th type of raw coal in the j-th coal blending scheme during the t-th period. i m is the unit price of the i-th type of raw coal. t Where Q represents the total amount of coal to be blended, and Q is the penalty for coke quality deviation. Let k be the measured value of the k-th quality indicator. Let λ be the target value of the k-th quality indicator. k Let ω1 and ω2 be the weights corresponding to the k-th quality indicator, and ω1 and ω2 be the weight coefficients of coal blending cost and coke quality, respectively. The constraints are: I it =I i,t-1 +y it ·s it -x ijt ·m t ≥I min Among them, a ik Let b be the k-th index value of the i-th type of raw coal. k I represents the upper or lower limit of the value of the k-th indicator. it Let s be the inventory of the i-th type of raw coal in period t. it Let y be the purchase quantity of the i-th type of raw coal in period t. it I is an indicator variable for whether the i-th type of raw coal was purchased. min Safety stock level; Based on inventory information and coal quality testing data, N feasible coal blending schemes that satisfy the objective function and constraints are randomly generated; for each feasible coal blending scheme, the proportion and type of raw coal are iteratively optimized through a genetic algorithm to satisfy the constraints. After each iteration, the fitness of the feasible coal blending scheme is calculated based on market prices and inventory information, combined with the objective function. After multiple rounds of iterative optimization, at least one feasible coal blending scheme with a fitness value greater than the preset value was determined as the coal blending scheme.

[0151] Furthermore, when the coal blending scheme optimization module 205 uses the expert-guided model to determine optimization suggestions based on the coal blending scheme, it is specifically used for: The application of expert-guided models analyzes the coal quality testing data of raw coal to determine the quality index of the raw coal; The effects of the characteristics of each type of raw coal on coke quality were analyzed and determined. Raw coal with a quality index below the index threshold and / or that would reduce coke quality was screened as problematic raw coal. Obtain alternative raw coal that can replace the problematic raw coal, and the quality index of the alternative raw coal is not lower than the index threshold and / or will improve the quality of coke. The problematic raw coal is replaced with a substitute raw coal to obtain a variety of new raw coals for blending. An additional blending scheme is generated based on the new raw coals for blending, and a suggestion for changing coal types is generated based on the substitute raw coals and the additional blending scheme.

[0152] Furthermore, the market forecasting model application module 206 is specifically used for: Causality tests were applied to quantify the causal influence of multiple exogenous variables on raw coal prices; Obtain historical price data for each type of raw coal; The time series features of historical price data and the causal influence strength of corresponding exogenous variables are obtained, and the time series features are standardized to obtain a dataset consisting of historical price data, its corresponding time series features, and Granger causal variable data. Build an LSTM model, use the dataset to train the LSTM model, and obtain the trained LSTM market price prediction model. Obtain real-time Granger causal variable data and time series features corresponding to market prices; The market price of raw coal, the time series characteristics of the market price, and real-time Granger causal variable data are input into the trained LSTM market price prediction model to predict the price at multiple future moments within a preset time period. The price trend is obtained based on the predicted price at each future moment.

[0153] Furthermore, when generating procurement plan suggestions corresponding to the coal blending scheme based on the price trend and inventory information of raw coal, the market forecasting model application module 206 is specifically used for: The price fluctuation coefficient is determined based on the price trend of raw coal. Extract the proportion, quality fluctuation coefficient, and planned consumption of each type of raw coal in the coal blending scheme; Calculate the lower limit of safe inventory and the upper limit of inventory warning for raw coal. The lower limit of safe inventory = average daily consumption × (procurement cycle + buffer days) × (1 + quality fluctuation coefficient) × (1 + price fluctuation coefficient). The upper limit of inventory warning = lower limit of safe inventory × (1 + price fluctuation buffer ratio). When the inventory information is less than the preset percentage of the safety stock limit, a shortage warning message is generated, and the purchase quantity is determined as (planned consumption - inventory) × (1 + emergency buffer ratio). When the inventory information is between the lower limit of safety stock and the upper limit of inventory warning, an early warning message is generated, and the purchase quantity is determined as (purchase cycle × average daily consumption) × fluctuation coefficient. When inventory information is greater than or equal to the inventory warning limit, a safety alert is generated, and the best time to purchase is determined by price trends.

[0154] Furthermore, the intelligent prediction system for coal feeding in coking plants 200 also includes: The coke quality determination module is used to determine the coke quality indicators for each coal blending scheme by applying an expert-guided model. The cost price determination module is used to calculate the predicted cost price corresponding to the coal blending scheme based on the predicted price of each raw coal. The sorting module is used to calculate the total evaluation score by assigning dynamic weights to the quality indicators and predicted cost prices of each coal blending scheme, and generate an evaluation priority sequence by sorting the total evaluation scores in descending order.

[0155] The sorting module is specifically used for: The causal influence of exogenous variables on raw coal prices is normalized, and a static causal weight vector is output. The exogenous variables and historical price data are spliced ​​together in time sequence to form the time series data of the exogenous variables; Construct an LSTM dynamic weight allocation model, add a fully connected layer before the LSTM dynamic weight allocation model, and use the static causal weight vector as the learnable initial weights. The LSTM dynamic weight allocation model is applied, and the time series data of exogenous variables are input, and the dynamic weight coefficients are output. The dynamic weights are obtained by multiplying the static causal weight vector and the dynamic weight coefficients element by element.

[0156] Furthermore, the intelligent prediction system for coal feeding in coking plants 200 also includes: The historical price acquisition module is used to obtain historical price data for each type of raw coal. The coal price volatility calculation module is used to calculate coal price volatility based on historical price data. The option premium conversion module is used to convert coal price volatility into option premium based on the option pricing model; The probability distribution generation module is used to generate the probability distribution path of coal prices through Monte Carlo simulation based on historical price data and coal price volatility. The optimal hedging ratio calculation module is used to calculate the optimal hedging ratio between futures and spot prices by combining the coal price distribution path and option premium. The hedging operation execution module is used to perform dynamic hedging operations based on the optimal hedging ratio.

[0157] Furthermore, the real-time data acquisition module 202 is specifically used for: Real-time coal quality testing data is obtained through the production management system; Use web crawling technology to capture real-time market prices; The system acquires dynamic scanning data of coal piles in the raw coal warehouse, combines it with computer vision algorithms to identify the volume of coal piles, and determines the first inventory data in the raw coal warehouse. Obtain the location and coal transport volume of vehicles awaiting unloading and vehicles en route; Predict the time when new raw coal will be delivered to the warehouse based on vehicle location; Based on the time of raw coal entering the warehouse and the amount of coal transported on the vehicles entering the warehouse, calculate the real-time warehouse entry data, add the real-time warehouse entry data to the first inventory data, and calculate the real-time inventory information. Based on vehicle location, predict the time of new raw coal delivery to the warehouse, including: Get vehicle speed; Determine the distance between the vehicle's location and the unloading endpoint; The transportation time is calculated based on vehicle speed and distance, and the raw coal storage time is calculated based on transportation time and floating time. If each historical delivery time is less than the average delivery time, then the fluctuation time is zero. If the proportion of historical delivery time exceeding the average delivery time reaches a first preset value, then calculate the difference between each historical delivery time and the average delivery time, calculate the mean of each difference to obtain the mean difference, obtain the vehicle's delivery delay rate within the preset time period, and calculate the floating time = mean difference × (1 + delivery delay rate / 100%).

[0158] The various variations and specific examples of the methods in the foregoing embodiments are also applicable to the intelligent prediction device for coking coal feed in this embodiment. Through the foregoing detailed description of the intelligent prediction method for coking coal feed, those skilled in the art can clearly understand the implementation method of the intelligent prediction device for coking coal feed in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.

[0159] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

[0160] Additionally, it should be understood that relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

Claims

1. A method for intelligent prediction of coal feed into coking plants, characterized in that, include: Obtain quality and cost indicators for various raw coals and cokes used in coal blending; Real-time access to coal quality testing data, market prices, and inventory information for each type of raw coal; The application of the expert-guided model is used to conduct in-depth analysis of the characteristics of the raw coal based on the coal quality test data, determine the analysis results of the raw coal, and determine the constraints based on the raw coal, the coke quality indicators, and the cost indicators. Based on the coal quality testing data, market price, and inventory information of each type of raw coal, a preliminary coal blending scheme is generated by applying a mathematical calculation model using mixed integer rules and genetic algorithms, with the goal of minimizing coke quality indicators and costs, and in conjunction with the constraints. Based on multiple interactions between the mathematical calculation model and the expert guidance model, the coal blending scheme is adjusted, and the expert guidance model determines optimization suggestions based on the coal blending scheme. A market forecasting model based on LSTM neural network and time-series causal analysis is applied to predict the price trend of raw coal in the future within a preset time period. Based on the price trend of raw coal and the inventory information, a procurement plan suggestion corresponding to the coal blending scheme is generated. Generate a coal intake evaluation report that includes at least the raw coal analysis results, the coal blending scheme, the scheme optimization suggestions, and the procurement scheme suggestions.

2. The method according to claim 1, characterized in that, Based on the coal quality testing data, market price, and inventory information for each type of raw coal, a mathematical calculation model employing mixed integer rules and a genetic algorithm is used to generate a coal blending scheme with the goal of minimizing coke quality indicators and costs. This scheme is combined with an expert-guided model based on the coal blending constraints provided by the raw coal. Construct the objective function using mixed-integer programming; The objective function is: minF=ω1C+ω2Q C=∑ i,t x ijt ·p i ·m t Where C is the coal blending cost, x ijt Let p be the proportion of the i-th type of raw coal in the j-th coal blending scheme during the t-th period. i m is the unit price of the i-th type of raw coal. t Where Q represents the total amount of coal to be blended, and Q is the penalty for coke quality deviation. Let k be the measured value of the k-th quality indicator. Let λ be the target value of the k-th quality indicator. k Let ω1 and ω2 be the weights corresponding to the k-th quality indicator, and ω1 and ω2 be the weight coefficients of coal blending cost and coke quality, respectively. The constraints are as follows: I it =I i,t-1 +y it ·s it -x ijt ·m t ≥I min Among them, a ik Let b be the k-th index value of the i-th type of raw coal. k I represents the upper or lower limit of the value of the k-th indicator. it Let s be the inventory of the i-th type of raw coal in period t. it Let y be the purchase quantity of the i-th type of raw coal in period t. it I is an indicator variable for whether the i-th type of raw coal was purchased. min Safety stock level; Based on the inventory information and coal quality testing data, N feasible coal blending schemes that satisfy the objective function and the constraints are randomly generated. For each feasible coal blending scheme, the proportions and types of raw coal are iteratively optimized using a genetic algorithm, while satisfying the aforementioned constraints. After each iteration, the fitness of the feasible coal blending scheme is calculated based on the market price and inventory information, combined with the objective function. After multiple iterations of optimization, at least one feasible coal blending scheme with a fitness greater than the preset value is determined as the coal blending scheme.

3. The method according to claim 1, characterized in that, The application, based on a market forecasting model using LSTM neural networks and time-series causal analysis, predicts the price trend of raw coal over a predetermined time period, including: Causality tests were applied to quantify the causal influence of multiple exogenous variables on raw coal prices; Obtain historical price data for each type of raw coal; The time series features of the historical price data and the causal influence strength of the corresponding exogenous variables are obtained, and the time series features are standardized to obtain a dataset consisting of historical price data, its corresponding time series features, and Granger causal variable data. Construct an LSTM model, train the LSTM model using a dataset, and obtain the trained LSTM market price prediction model. Obtain real-time Granger causal variable data and time series features corresponding to market prices; The market price of the raw coal, the time series characteristics corresponding to the market price, and the real-time Granger causal variable data are input into the trained LSTM market price prediction model to predict the price at multiple future moments within a preset future time period. The price trend is obtained based on the predicted price at each future moment.

4. The method according to claim 1, characterized in that, The step of generating a procurement plan suggestion corresponding to the coal blending scheme based on the price trend of raw coal and the inventory information includes: The price fluctuation coefficient is determined based on the stated price trend of raw coal; Extract the proportion, quality fluctuation coefficient, and planned consumption of each type of raw coal in the coal blending scheme; Calculate the lower limit of safe inventory and the upper limit of inventory warning for raw coal. The lower limit of safe inventory = average daily consumption × (procurement cycle + buffer days) × (1 + quality fluctuation coefficient) × (1 + price fluctuation coefficient). The upper limit of inventory warning = lower limit of safe inventory × (1 + price fluctuation buffer ratio). When the inventory information is less than the preset percentage of the safety stock lower limit, a shortage warning message is generated, and the purchase quantity is determined as (planned consumption - inventory) × (1 + emergency buffer ratio). When the inventory information is between the lower limit of the safety stock and the upper limit of the inventory warning, an early warning message is generated, and the purchase quantity is determined as (purchase cycle × average daily consumption) × fluctuation coefficient. When the inventory information is greater than or equal to the inventory warning limit, a safety alert is generated, and the optimal purchasing period is determined based on the price trend.

5. The method according to claim 3, characterized in that, If the coal blending scheme includes multiple schemes, the method further includes: The coke quality indicators for each coal blending scheme are determined using an expert-guided model. Calculate the predicted cost price corresponding to the coal blending scheme based on the predicted price of each type of raw coal; The quality indicators and predicted cost prices of each coal blending scheme are dynamically weighted to calculate the total evaluation score, and the evaluation priority sequence is generated by sorting the total evaluation scores in descending order.

6. The method according to claim 5, characterized in that, Before calculating the total evaluation score by assigning dynamic weights to the quality indicators and predicted cost prices of each coal blending scheme, the method further includes: The causal influence of exogenous variables on raw coal prices is normalized, and a static causal weight vector is output. The exogenous variables and historical price data are spliced ​​together in time sequence to form the time series data of the exogenous variables; Construct an LSTM dynamic weight allocation model, add a fully connected layer before the LSTM dynamic weight allocation model, and use the static causal weight vector as the learnable initial weights. The LSTM dynamic weight allocation model is applied, with exogenous variable time series data as input, and dynamic weight coefficients as output. The dynamic weights are obtained by multiplying the static causal weight vector and the dynamic weight coefficients element by element.

7. The method according to claim 1, characterized in that, The optimization suggestions include suggestions for changing the coal type. The expert guidance model determines the optimization suggestions based on the coal blending scheme, including: The expert-guided model is used to analyze the coal quality test data of the raw coal to determine the quality index of the raw coal; The effects of the characteristics of each type of raw coal on coke quality were analyzed and determined. Raw coal with a quality index below the index threshold and / or that would reduce coke quality was screened as problematic raw coal. Obtain alternative raw coal that can replace the problematic raw coal, wherein the quality index of the alternative raw coal is not lower than the index threshold and / or improves the quality of coke; The problematic raw coal is replaced by the alternative raw coal to obtain a variety of new raw coals for blending. An additional blending scheme is generated based on the new raw coals for blending. A suggestion for changing coal types is generated based on the alternative raw coals and the additional blending scheme.

8. The method according to claim 1, characterized in that, The method further includes: Obtain historical price data for each type of raw coal; Calculate the coal price volatility based on the historical price data; Based on the option pricing model, the coal price volatility is converted into an option premium; Based on the historical price data and the coal price volatility, a probability distribution path for coal prices is generated through Monte Carlo simulation. Based on the aforementioned coal price distribution path and option premium, calculate the optimal hedging ratio between futures and spot prices; Perform dynamic hedging operations based on the optimal hedging ratio.

9. The method according to claim 1, characterized in that, The real-time acquisition of coal quality testing data, market prices, and inventory information for each type of raw coal includes: Real-time coal quality testing data is obtained through the production management system; Use web crawling technology to capture real-time market prices; The system acquires dynamic scanning data of coal piles in the raw coal warehouse, combines it with computer vision algorithms to identify the volume of coal piles, and determines the first inventory data in the raw coal warehouse. Obtain the location and coal transport volume of vehicles awaiting unloading and vehicles en route; Based on the vehicle's location, the time when new raw coal is delivered to the warehouse is predicted; Based on the raw coal's arrival time and the amount of coal transported on the vehicles, real-time warehousing data is calculated. The real-time warehousing data is then added to the first inventory data to obtain real-time inventory information. The prediction of the new coal arrival time based on the vehicle location includes: Get vehicle speed; Determine the distance between the vehicle's location and the unloading endpoint; The transportation time is calculated based on the vehicle speed and the distance, and the raw coal storage time is calculated based on the transportation time and the floating time. If each historical delivery time is less than the average delivery time, then the fluctuation time is zero. If the proportion of historical delivery time exceeding the average delivery time reaches a first preset value, then calculate the difference between each historical delivery time and the average delivery time, calculate the mean of each difference to obtain the mean difference, obtain the vehicle's delivery delay rate within the preset time period, and calculate the floating time = mean difference × (1 + delivery delay rate / 100%).

10. A coking coal feed intelligent prediction system, characterized in that, include: The basic data acquisition module is used to acquire various raw material coal and coke quality and cost indicators for coal blending; The real-time data acquisition module is used to acquire coal quality testing data, market prices, and inventory information for each type of raw coal in real time. The constraint determination module is used to apply an expert-guided model to conduct in-depth analysis of the characteristics of raw coal, determine the analysis results of raw coal, and determine the constraint conditions based on the raw coal, the coke quality index, and the cost index. The preliminary coal blending scheme generation module is used to generate a preliminary coal blending scheme based on the coal quality test data, market price, and inventory information of each raw coal, using a mathematical calculation model with mixed integer rules and genetic algorithms, with the goal of minimizing coke quality indicators and costs, and in combination with the constraints. The coal blending scheme optimization module is used to adjust the coal blending scheme based on multiple interactions between the mathematical calculation model and the expert guidance model. The expert guidance model determines the scheme optimization suggestions based on the coal blending scheme. The market forecasting model application module is used to apply a market forecasting model based on LSTM neural network and time-series causal analysis to predict the price trend of raw coal in the future within a preset time period, and generate procurement plan suggestions corresponding to the coal blending plan based on the price trend of raw coal and the inventory information. The coal intake evaluation report generation module is used to generate a coal intake evaluation report that includes at least the raw coal analysis results, the coal blending scheme, the scheme optimization suggestions, and the procurement scheme suggestions.

Images