Adaptive Control Method and System for Coking Coal Blending Model
By using adaptive control methods and systems, combined with multi-source data and small coke oven experiments, model parameters are dynamically corrected, implicit strategies are discovered, and optimized coking coal blending schemes are generated. This solves the adaptability and cost problems of existing models and achieves high-precision and low-cost coking production.
Patent Information
- Application Number
- CN202611365262.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-04
- Publication Date
- 2026-11-17
AI Technical Summary
Existing coking coal blending prediction models are unable to adapt to fluctuations in raw coal quality and changes in coke oven operating conditions, resulting in decreased prediction accuracy. Furthermore, they cannot effectively utilize implicit coal blending experience, leading to high trial-and-error costs and fragmented data.
By collecting multi-source production data, an adaptive coke quality prediction model is constructed. Feedback adjustments are made by combining experimental data from small coke ovens to uncover implicit coal blending strategies, a knowledge base for coal type replacement paths is built, and a multi-objective optimization algorithm is used to generate recommended coal blending schemes.
This improved the long-term adaptability of the prediction model, reduced trial-and-error costs, generated a more stable and lower-cost coal blending scheme, and ensured stable coke quality.
Smart Images

Figure CN123120085A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of coking coal blending technology, and in particular to an adaptive control method and system for coking coal blending models. Background Technology
[0002] Coking coal blending is a core step in the coking production process. By blending various types of coal with different properties in specific proportions, coal resources can be utilized rationally, and the quality of the produced coke can be ensured to meet the requirements of blast furnace smelting. With the development of industrial informatization and intelligentization, using mathematical models and computer technology for coal blending prediction and scheme optimization has become an important means to improve coking quality and reduce production costs.
[0003] Existing methods for coal blending prediction and optimization typically rely on fixed statistical regression or machine learning prediction models based on historical production data to predict the mechanical and thermal strength of coke. When generating a coal blending scheme, the optimization algorithm, based on feedback from the prediction model, seeks the lowest-cost blend ratio under set constraints. Simultaneously, to verify the feasibility of new coal types or formulas, small-scale coke oven experiments are usually conducted before large-scale industrial production, serving as a reference for large-scale industrial production.
[0004] However, existing coal blending prediction models are often static and struggle to adapt to dynamic disturbances such as fluctuations in raw coal quality and changes in coke oven operating conditions, leading to a gradual decline in prediction accuracy over time. Furthermore, while companies have accumulated a wealth of historically successful coal blending schemes, the underlying blending logic and coal type substitution patterns are often implicit. Existing static models cannot automatically uncover and utilize these implicit strategies, resulting in reliance on manual experience when generating new schemes, leading to high trial-and-error costs. Simultaneously, a scaling effect exists between small-scale coke oven experiments and large-scale industrial production. Experimental results from small coke ovens cannot directly and quantitatively feed back into and correct large-scale production prediction models, resulting in a disconnect between large and small coke oven data and hindering the formation of a closed-loop adaptive adjustment mechanism for model accuracy. Summary of the Invention
[0005] To address the aforementioned issues, this application provides an adaptive control method and system for coking coal blending models.
[0006] Firstly, this application provides an adaptive control method for coking coal blending models, employing the following technical solution: An adaptive control method for coking coal blending models includes the following steps: S1: Collect multi-source coal blending production data, which includes raw coal property data, historical coal blending scheme data, large-scale coke quality data, and small coke oven experimental coal blending and coke quality data. S2: Based on the historical coal blending scheme data and the corresponding raw coal property data and large-scale coke quality data, construct an initial version of the coke quality prediction model; S3: Based on the experimental coal blending and coke quality data of the small coke oven and the corresponding coke quality data of large-scale production, calculate the prediction deviation, and use the prediction deviation to adaptively adjust the initial version of the coke quality prediction model, update the model parameters, and obtain an adaptive coke quality prediction model. S4: Deconstruct multiple sets of historical excellent coal blending schemes, use association rule mining algorithms to identify implicit coal blending strategies, and build a knowledge base for coal type replacement paths; S5: Using the lowest cost and quality as multi-objective optimization functions, and combining the adaptive coke quality prediction model with the coal type replacement path knowledge base, the coal blending scheme is reconstructed and optimized to generate a recommended coal blending scheme.
[0007] By adopting the above technical solutions, the automatic aggregation of multi-source coal blending production data can be realized. The initial model provides basic coke quality prediction capabilities, and adaptive feedback adjustment is carried out by combining experimental data from small coke ovens. This significantly reduces the prediction deviation between large and small coke ovens and solves the technical problem that static models cannot adapt to fluctuations in the coal blending environment. At the same time, by deconstructing and mining strategies from historical excellent solutions, implicit coal blending experience can be transformed into explicit control rules, reducing the trial and error costs of coal blending solutions and generating more stable and lower-cost coal blending solutions.
[0008] Optionally, in step S1, the multi-source coal blending production data is integrated and cleaned using an ETL tool. The cleaning process includes removing outliers, imputing missing values, and aligning raw coal and coke data based on timestamps.
[0009] By adopting the above technical solution, ETL tools can efficiently integrate data from different systems (such as DCS, quality inspection systems, etc.), remove noise interference, and align the raw coal feeding time with the coking time based on timestamps, providing a high-quality data foundation for the accurate training of subsequent models.
[0010] Optionally, in step S3, the parameter adjustment formula for the adaptive feedback adjustment is: in, This is the updated model parameter matrix; This is the original model parameter matrix; The learning rate parameter; Data on actual coke quality from large-scale production; For model predicted values; The feature input matrix for raw coal; A correction factor for the collaborative production of small coke ovens at a large scale; This is the data on the quality value of coke from the small coke oven experiment.
[0011] By adopting the above technical solution, the actual deviation of large-scale production and the experimental deviation of small coke ovens are taken into account, effectively eliminating the scale effect between large and small coke ovens. This allows the prediction model to automatically update parameters as the production status changes, ensuring high prediction accuracy under long-term operation.
[0012] Optionally, in step S4, the specific steps for deconstructing multiple sets of historically excellent coal blending schemes are as follows: Historical coal blending schemes were grouped according to coke quality indicators, and schemes with comprehensive scores higher than a set threshold were selected as excellent historical coal blending schemes. Each set of historically excellent coal blending schemes is broken down into a raw coal ratio vector and a single coal quality index vector.
[0013] By adopting the above technical solution, and by setting a reasonable threshold to filter out high-performing historical schemes, and by structurally decomposing them into vector forms for mining, the interference of inefficient schemes can be effectively filtered out, thereby improving the targeting and accuracy of implicit strategy mining.
[0014] Optionally, the association rule mining algorithm uses the Apriori algorithm, by setting a support threshold. and confidence threshold By mining the co-occurrence relationships and proportional substitution constraints among different coal types in the raw coal blending vector, implicit coal blending strategies can be extracted.
[0015] By adopting the above technical solution, the Apriori algorithm can quickly discover strong correlation rules for coal type matching in a large number of historical schemes. For example, a certain type of strongly caking coal must be matched with a certain type of weakly caking coal in a specific ratio, thereby transforming the implicit experience of experts into specific quantitative rule constraints.
[0016] Optionally, in step S4, the steps for constructing the coal type replacement path knowledge base are as follows: Based on the discovered implicit coal blending strategy, the equivalence of different coal types under the same coke quality output is calculated, a confidence score is generated for replacing the first coal type with the second coal type, and coal type pairs with confidence scores higher than a preset replacement threshold are stored in the coal type replacement path knowledge base.
[0017] By adopting the above technical solution, the equivalence of mutual substitution between different coal types is quantitatively calculated, and reasonable coal type substitution paths are automatically generated. This makes it easier for the system to automatically recommend suitable alternative coal types when there is a shortage of a certain type of coal or when prices soar, thus ensuring stable production and optimizing procurement costs.
[0018] Optionally, in step S5, the multi-objective optimization objective function is expressed as:
[0019] The constraints include:
[0020] in, Let be the coal blending ratio vector to be solved; This is a function for calculating coal blending costs. The comprehensive coke quality score is calculated based on an adaptive coke quality prediction model. These are constraints set based on a knowledge base of coal type replacement paths and production process parameters, including restrictions on the proportion of a single type of coal, ash content restrictions, and sulfur content restrictions.
[0021] By adopting the above technical solution, a multi-objective optimization model was constructed, which regards cost control and quality control as two mutually restrictive objectives, and transforms the coal type replacement knowledge base into explicit constraints, ensuring that the optimization solution achieves the ultimate cost reduction while meeting quality standards.
[0022] Optionally, the non-dominated sorting genetic algorithm NSGA-II is used to iteratively solve the multi-objective optimization objective function to obtain the Pareto optimal solution set, and the target scheme is selected from the Pareto optimal solution set as the recommended coal blending scheme according to the actual production strategy.
[0023] By adopting the above technical solutions, the NSGA-II algorithm can efficiently search for the Pareto front in a multidimensional nonlinear space, avoid getting trapped in local optima, provide decision-makers with a variety of coal blending alternatives that take into account both quality and cost, and improve the flexibility of scheme generation.
[0024] Optionally, step S6 is also included: the recommended coal blending scheme is sent to the coal preparation and feeding system for industrial production, and the actual coke quality data obtained after production is collected in real time. The actual coke quality data is then fed back to step S3 to form a closed-loop adaptive feedback adjustment.
[0025] By adopting the above technical solution, each production result can be used as a new sample input to drive the model's self-evolution, thus achieving continuous improvement in coal blending quality.
[0026] Secondly, this application provides a system that applies an adaptive feedback adjustment and implicit strategy mining method based on a coking coal blending model, including: The data acquisition module is used to collect multi-source coal blending production data; The model building module is used to build the initial version of the coke quality prediction model. The feedback adjustment module is used to calculate the prediction deviation and use experimental data from small coke ovens to adaptively adjust the model parameters and update and generate an adaptive coke quality prediction model. The strategy mining module is used to deconstruct historical excellent coal blending schemes and mine implicit coal blending strategies to build a knowledge base for coal type replacement paths. The scheme reconstruction module is used to reconstruct and optimize the coal blending scheme by combining the adaptive coke quality prediction model and the coal type replacement path knowledge base, and generate a recommended coal blending scheme.
[0027] By adopting the above technical solution and decoupling the functional modules, a highly integrated and fast-response intelligent coal blending hardware / software interaction system was constructed, ensuring the smooth implementation and operation of the algorithm in a large-scale industrial production environment.
[0028] In summary, this application includes at least one of the following beneficial technical effects: By using an adaptive feedback adjustment mechanism, experimental data from small coke ovens are fused and calibrated with large-scale production data. The model parameters are then dynamically corrected based on actual deviations in large-scale production. This results in significantly improved prediction accuracy and the ability to adapt to fluctuations in raw coal quality over the long term. The prediction error for coke thermal intensity is successfully controlled within an extremely low range. By utilizing implicit strategy mining techniques based on association rule mining algorithms, a knowledge base for coal type replacement paths was constructed. Implicit coal blending experience was transformed into explicit rule constraints and equivalent replacement paths, which achieved the effect of guiding the procurement and replacement of new coal types, thereby shortening the experimental cycle of small coke ovens and effectively reducing the cost of coal blending trial and error. Attached Figure Description
[0029] Figure 1 This is a flowchart of the adaptive control method for coking coal blending models.
[0030] Figure 2 This is a schematic diagram of an adaptive feedback adjustment mechanism.
[0031] Figure 3 This is a flowchart of the reconstructing of coal blending schemes based on multiple objectives.
[0032] Figure 4 This is a schematic diagram of the adaptive control system of the coking coal blending model. Detailed Implementation
[0033] The present application will be further described in detail below with reference to all the accompanying drawings.
[0034] This application discloses an adaptive control method for a coking coal blending model.
[0035] Reference Figures 1 to 3 An adaptive control method for coking coal blending models includes the following steps: S1: Collect multi-source coal blending production data, including raw coal property data, historical coal blending scheme data, large-scale coke quality data, and experimental coal blending and coke quality data from small coke ovens. In this step, the hardware acquisition terminal establishes communication connections with the coking plant's bulk material system, DCS (Distributed Control System), quality inspection system, and LIMS laboratory management system. Raw coal property data includes the raw coal's ash content (e.g., low-ash coal with an ash content Ad of 7.5% or high-ash coal with an ash content Ad of 10.5%), volatile matter (e.g., volatile matter Vdaf of 20.0% or 32.0%), total sulfur (St,d), caking index (G value), and maximum thickness of the plastic layer (Y value), etc. Large-scale coke quality data includes the coke's mechanical strength (M40, M10) and thermal strength indices (reactivity CRI, post-reaction strength CSR). Multi-source coal blending production data is integrated and cleaned using ETL tools. The cleaning process includes removing outliers caused by sensor malfunctions, imputing missing values using averaging or interpolation, and aligning the time of raw coal entering the blending silo with the time of coking based on timestamps to ensure consistency of "raw materials-process-quality" in the time dimension.
[0036] S2: Based on historical coal blending scheme data, corresponding raw coal property data, and large-scale coke quality data, a preliminary coke quality prediction model is constructed. This model is built using machine learning algorithms such as multiple linear regression or BP neural network. Its input features are the proportion of each type of coal in the coal blending scheme and the corresponding coal quality index of each type of coal. The output features are the predicted coke quality indicators (M40, M10, CRI, CSR).
[0037] S3: Based on the experimental coal blending and coke quality data from small coke ovens and the corresponding coke quality data from large-scale production, the prediction deviation is calculated. This deviation is then used to adaptively adjust the initial coke quality prediction model using feedback, updating the model parameters to obtain an adaptive coke quality prediction model. The parameter adjustment formula for adaptive feedback adjustment is:
[0038] middle, This is the updated model parameter matrix; This is the original model parameter matrix; The learning rate parameter; Data on actual coke quality from large-scale production; For model predicted values; The feature input matrix for raw coal; A correction factor for the collaborative production of small coke ovens at a large scale; This provides data on coke quality from experiments conducted in small coke ovens. Using this formula, the model can adaptively eliminate the scaling effect between large and small coke ovens.
[0039] S4: Deconstruct multiple historical excellent coal blending schemes, and use an association rule mining algorithm to identify implicit coal blending strategies, constructing a knowledge base for coal type replacement paths. The specific deconstruction steps are as follows: First, group historical coal blending schemes according to coke quality indicators, and select schemes with a comprehensive score higher than a set threshold (e.g., a comprehensive score greater than 85 points) as historical excellent coal blending schemes; then, decompose each group of historical excellent coal blending schemes into raw coal ratio vectors and single-coal quality indicator vectors. The association rule mining algorithm uses the Apriori algorithm (in other embodiments, the FP-Growth algorithm can also be used), by setting a support threshold. (e.g., 0.2 or 0.3) and confidence threshold (e.g., 0.7 or 0.8) Mine the co-occurrence relationships and proportional substitution constraints among different coal types in the raw coal blending vector. Based on the mined implicit coal blending strategies, calculate the equivalence of different coal types under the same coke quality output, generate confidence scores for replacing the first coal type (e.g., gas coal) with the second coal type (e.g., 1 / 3 coking coal), and store coal type pairs with confidence scores higher than the preset substitution threshold into the coal type substitution path knowledge base.
[0040] S5: Using the lowest cost and quality as multi-objective optimization functions, and combining an adaptive coke quality prediction model with a coal type replacement path knowledge base, the coal blending scheme is reconstructed and optimized to generate a recommended coal blending scheme. The multi-objective optimization objective function is expressed as:
[0041] The constraints include:
[0042] in, Let be the coal blending ratio vector to be solved; This is a function for calculating coal blending costs. The comprehensive coke quality score is calculated based on an adaptive coke quality prediction model. The constraints are set based on the coal type replacement path knowledge base and production process parameters, including single coal type ratio limits, ash content limits, and sulfur content limits.
[0043] In this embodiment, the non-dominated sorting genetic algorithm NSGA-II (or the multi-objective particle swarm optimization algorithm MOPSO in other embodiments) is used to iteratively solve the multi-objective optimization objective function to obtain the Pareto optimal solution set, and the target scheme is selected from the Pareto optimal solution set as the recommended coal blending scheme according to the actual production strategy.
[0044] The method in this embodiment also includes S6: sending the recommended coal blending scheme to the PLC controller of the coal preparation and feeding system through the communication interface, controlling the feeding belt scales of each coal blending silo to carry out industrial production, and collecting the actual coke quality data obtained after production in real time, and feeding the actual coke quality data back to S3 to form a closed-loop adaptive feedback adjustment.
[0045] The working principle of the adaptive control method for coking coal blending model in this application embodiment is as follows: Data silos are broken down using ETL tools to collect and align multi-source heterogeneous coal blending production data from the field, establishing an initial prediction model based on this. In actual operation, considering coke oven aging, coal quality fluctuations, and the scale effect between large and small coke ovens, this embodiment designs an adaptive update formula based on dual feedback from prediction bias and small coke oven experimental data. This dynamically corrects the weight parameters of the prediction model, enabling it to self-evolve. Furthermore, by performing vector deconstruction on historical best practices, association rule mining algorithms (such as the Apriori algorithm) are used to automatically identify coal blending rules hidden in historical experience (such as coal type co-occurrence ratios and equivalent replacement patterns), establishing a coal type replacement path knowledge base. Finally, using the adaptive model and replacement knowledge base as input and constraints, the NSGA-II multi-objective optimization algorithm is employed to find the optimal balance between cost and quality, generating recommended schemes that can directly guide industrial production. Through closed-loop feedback, the scheme is continuously iterated, thereby achieving the goal of stabilizing coke quality and reducing coal blending costs.
[0046] This application also discloses an adaptive control system for a coking coal blending model.
[0047] Reference Figure 4 An adaptive control system for coking coal blending models, comprising: The data acquisition module is used to collect multi-source coal blending production data. In this embodiment, the data acquisition module interfaces with multiple underlying information systems of the coking plant via industrial Ethernet. Specifically, it includes interface units that connect to the bulk material system interface, DCS data interface, LIMS quality inspection system interface, and coal preparation and feeding system interface, respectively. The data acquisition module not only collects real-time raw coal property data (such as ash content, volatile matter, and caking index), historical coal blending scheme data, and large-scale coke quality data, but also specifically collects experimental coal blending and coke quality data conducted on a 40kg small coke oven in the laboratory. The collected multi-source data is cleaned internally by an ETL unit to remove noise and is aligned based on timestamps before being stored in the system's built-in basic database.
[0048] The model building module is used to construct an initial version of the coke quality prediction model. This module includes a model training unit, which calls historical coal blending scheme data, corresponding raw coal property data, and large-scale coke quality data, and uses a BP neural network algorithm for offline training to establish an initial nonlinear mapping relationship model between various indicators of blended coal and coke M40, M10, CRI, and CSR.
[0049] The feedback adjustment module calculates prediction deviations and uses experimental data from the small coke oven to adaptively adjust model parameters, updating and generating an adaptive coke quality prediction model. This module is the core self-learning unit of the system; its input is connected to the data acquisition module and the small coke oven data interface, and its output is connected to the adaptive model memory. The feedback adjustment module internally includes a deviation calculation unit and a parameter update unit. The deviation calculation unit calculates the actual values from large-scale production in real time. Compared with model predictions The deviation, and the experimental values of the small coke oven Compared with actual production value The deviation is used to dynamically adjust the neural network weight matrix W of the prediction model according to the parameter adjustment formula in step S3, thereby enabling the model to respond dynamically and agilely to changes in production conditions.
[0050] The strategy mining module is used to deconstruct historical excellent coal blending schemes and mine implicit coal blending strategies, constructing a coal type replacement path knowledge base. This module includes a scheme deconstruction unit and an association rule mining unit. The scheme deconstruction unit selects historical excellent schemes corresponding to high-quality coke based on quality inspection data and decomposes them into blending ratio vectors. The association rule mining unit uses the Apriori algorithm to mine implicit rules for single coal blending from these vectors and calculates the equivalent replacement confidence between different coal types. The mined strategies and replacement paths are structured and stored in the coal type replacement path knowledge base.
[0051] The scheme reconstruction module combines an adaptive coke quality prediction model with a coal type replacement path knowledge base to reconstruct and optimize coal blending schemes, generating recommended coal blending schemes. This module includes an optimization solver that calls the NSGA-II algorithm (Non-Dominated Sorting Genetic Algorithm II). The scheme reconstruction module uses minimum cost and quality compliance as objective functions, employs the adaptive coke quality prediction model as the quality constraint function, and uses the replacement rules in the coal type replacement path knowledge base as boundary constraints for decision variables. Through iterative calculation, it generates a Pareto optimal solution set and selects the optimal scheme from the solution set as the recommended coal blending scheme based on a preset operating strategy (such as cost priority or quality priority). This recommended scheme is sent to the PLC of the coal preparation and feeding system via a communication interface, enabling automatic execution and closed-loop control of the scheme. 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.
Claims
1. A self-adaptive regulation method for a coking coal blending model, characterized in that, Includes the following steps: S1: Collect multi-source coal blending production data, which includes raw coal property data, historical coal blending scheme data, large-scale coke quality data, and small coke oven experimental coal blending and coke quality data. S2: Based on the historical coal blending scheme data and the corresponding raw coal property data and large-scale coke quality data, construct an initial version of the coke quality prediction model; S3: Based on the experimental coal blending and coke quality data of the small coke oven and the corresponding coke quality data of large-scale production, calculate the prediction deviation, and use the prediction deviation to adaptively adjust the initial version of the coke quality prediction model, update the model parameters, and obtain an adaptive coke quality prediction model. S4: Deconstruct multiple sets of historical excellent coal blending schemes, use association rule mining algorithms to identify implicit coal blending strategies, and build a knowledge base for coal type replacement paths; S5: Using the lowest cost and quality as multi-objective optimization functions, and combining the adaptive coke quality prediction model with the coal type replacement path knowledge base, the coal blending scheme is reconstructed and optimized to generate a recommended coal blending scheme.
2. The coking coal blending model self-adaptive regulation method according to claim 1, characterized in that, In step S1, the multi-source coal blending production data is integrated and cleaned using an ETL tool. The cleaning process includes removing outliers, imputing missing values, and aligning raw coal and coke data based on timestamps.
3. The coking coal blending model self-adaptive regulation method according to claim 1, characterized in that, In step S3, the parameter adjustment formula for the adaptive feedback adjustment is: ; wherein, is the updated model parameter matrix; is the original model parameter matrix; is the learning rate parameter; is the actual value data of large production coke quality; is the model predicted value data; is the characteristic input matrix of raw coal; is the small coke oven large production collaborative correction factor; is the small coke oven experimental coke quality value data.
4. The coking coal blending model self-adaptive regulation method according to claim 1, characterized in that, In step S4, the specific steps for deconstructing multiple historically excellent coal blending schemes are as follows: Historical coal blending schemes were grouped according to coke quality indicators, and schemes with comprehensive scores higher than a set threshold were selected as excellent historical coal blending schemes. Each set of historically excellent coal blending schemes is broken down into a raw coal ratio vector and a single coal quality index vector.
5. The coking coal blending model self-adaptive regulation method according to claim 4, characterized in that, The association rule mining algorithm adopts an Apriori algorithm, and by setting a support threshold and a confidence threshold , the co-occurrence relationship and the proportional replacement constraint among different coal types in the raw coal blending ratio vector are mined, and the implicit coal blending strategy is extracted.
6. The coking coal blending model self-adaptive regulation method according to claim 5, characterized in that, In step S4, the steps for constructing the coal type replacement path knowledge base are as follows: Based on the discovered implicit coal blending strategy, the equivalence of different coal types under the same coke quality output is calculated, a confidence score is generated for replacing the first coal type with the second coal type, and coal type pairs with confidence scores higher than a preset replacement threshold are stored in the coal type replacement path knowledge base.
7. The coking coal blending model self-adaptive regulation method according to claim 1, characterized in that, In step S5, the multi-objective optimization objective function is expressed as: ; The constraints include: ; wherein, is a coal blending ratio vector to be solved; is a coal blending cost calculation function; is a coke quality comprehensive score calculated based on an adaptive coke quality prediction model; is a constraint condition based on a coal replacement path knowledge base and production process parameter setting, including single coal ratio limit, ash limit, and sulfur limit.
8. The coking coal blending model self-adaptive regulation method according to claim 7, characterized in that, The non-dominated sorting genetic algorithm NSGA-II is used to iteratively solve the multi-objective optimization objective function to obtain the Pareto optimal solution set, and the target scheme is selected from the Pareto optimal solution set as the recommended coal blending scheme according to the actual production strategy.
9. The coking coal blending model self-adaptive regulation method according to claim 1, characterized in that, It also includes step S6: sending the recommended coal blending scheme to the coal preparation and feeding system for industrial production, and collecting the actual coke quality data obtained after production in real time, and feeding the actual coke quality data back to step S3 to form a closed-loop adaptive feedback adjustment.
10. A system for applying the self-adaptive control method of the coking coal blending model according to any one of claims 1 to 9, characterized in that, include: The data acquisition module is used to collect multi-source coal blending production data; The model building module is used to build the initial version of the coke quality prediction model. The feedback adjustment module is used to calculate the prediction deviation and use experimental data from small coke ovens to adaptively adjust the model parameters and update and generate an adaptive coke quality prediction model. The strategy mining module is used to deconstruct historical excellent coal blending schemes and mine implicit coal blending strategies to build a knowledge base for coal type replacement paths. The scheme reconstruction module is used to reconstruct and optimize the coal blending scheme by combining the adaptive coke quality prediction model and the coal type replacement path knowledge base, and generate a recommended coal blending scheme.
