Ore grinding system control method based on large model and cuckoo algorithm

Abstract

The invention relates to the technical field of industrial automation, and discloses an ore grinding system control method based on a large model and a cuckoo algorithm. The method comprises the following steps: setting parameters of a cuckoo algorithm, and generating candidate solutions based on control parameters of an ore grinding system; performing fitness scoring on the candidate solutions through a large model, and recording fitness corresponding to all the candidate solutions; adding global disturbance to the candidate solutions in combination with a Levy flight mechanism, recalculating the fitness, performing an elimination mechanism based on fitness scores until a preset number of times is reached, and recording the candidate solution corresponding to the optimal fitness score in the two fitness scoring processes; and the obtained candidate solutions are applied to the ore grinding system, actual feedback is collected, the deviation value is calculated in combination with the large model, the candidate solutions are updated based on the deviation value, and the second step is returned to conduct fitness scoring on all the candidate solutions again till the ore grinding task is completed. The problem that an algorithm model adopted by an existing ore grinding system control method cannot meet the industrial control requirement is solved.

Description

Technical Field

[0001] This invention relates to the field of industrial automation technology, and in particular to a control method for applying large model and cuckoo algorithm to grinding systems. Background Technology

[0002] In actual production, the operation of grinding systems involves several key process parameters, including feed rate, ball rate, water feed rate, and discharge concentration. These parameters exhibit significant nonlinear coupling relationships and are subject to various uncertainties such as ore properties, equipment status, and environmental disturbances, resulting in highly dynamic and complex systems. For a long time, traditional grinding control systems have relied primarily on manual experience or rule-based control methods, which have limitations in responding to complex operating conditions. Even though some mineral processing enterprises have introduced automated monitoring and process control systems, achieving real-time parameter acquisition and preliminary adjustment, existing systems still fall short in terms of intelligent and refined management due to insufficient core modeling and optimization capabilities. Specifically, the current optimization and control of grinding systems mainly face the following problems: Traditional mechanistic models struggle to adequately describe the multivariable, strongly coupled, and nonlinear dynamic characteristics of the grinding process. High parameter uncertainty often leads to model mismatch. They are highly dependent on the quality of historical data and the representativeness of the samples, and have limited generalization ability for extreme or novel operating conditions, resulting in limited modeling capabilities.

[0003] Existing optimization methods are mostly single-objective or simple multi-objective optimizations. Commonly used genetic algorithms and particle swarm optimization are prone to getting trapped in local optima in high-dimensional constrained spaces, which limits their optimization ability and convergence speed. They are difficult to meet the global optimization needs of complex industrial scenarios, resulting in limitations of optimization algorithms.

[0004] The optimization and control processes are mostly open-loop or semi-closed-loop, failing to fully utilize actual operational feedback for adaptive parameter adjustment. As a result, the system struggles to achieve long-term high efficiency and dynamic stability, leading to a lack of feedback and adaptive mechanisms.

[0005] Existing systems lack effective adaptive compensation and fault tolerance mechanisms for abnormal situations such as sensor failure, model distortion, and extreme disturbances, which can easily lead to a sharp drop in system efficiency or even instability, affecting production continuity and process safety, and have weak robustness to abnormal and extreme operating conditions.

[0006] Existing methods often separate parameter optimization from process control, lacking closed-loop coordination between model evaluation, parameter optimization, and actual execution. This makes it difficult to effectively implement optimization results, hindering the intelligent upgrading of the production process and resulting in insufficient integration of optimization and control.

[0007] Therefore, there is an urgent need for a grinding system control method to meet the above-mentioned industrial control requirements. Summary of the Invention

[0008] This invention provides a grinding system control method based on a large model and the cuckoo algorithm to solve the problem that the algorithm models used in existing grinding system control methods cannot meet the requirements of industrial control.

[0009] To achieve the above objectives, the present invention employs the following technical solution: In a first aspect, the present invention provides a grinding system control method based on a large model and the cuckoo algorithm, comprising the following steps: Step 1: Set the parameters of the Cuckoo Algorithm and generate candidate solutions based on the control parameters of the grinding system; The generated candidate solutions are a set of grinding system control parameters. Step 2: Use the large model to score the fitness of all candidate solutions and record the fitness of each candidate solution; In step 2, the large-scale model used is the industrial-grade large-scale model. The industrial-grade large-scale model comprehensively models the multivariable, nonlinear, and strongly coupled processes of the grinding system, realizes the accurate evaluation of the efficiency of parameter combinations under different operating conditions, and provides a scientific and reliable objective function basis for optimization.

[0010] Step 3: Combine the Levy flight mechanism to add global perturbation to all candidate solutions, re-execute step 2, and perform the elimination mechanism based on fitness score. Repeat step 2 until the preset number of times is reached, and record the candidate solution corresponding to the best fitness score in the two fitness scoring processes. Step 4: Apply the candidate solutions obtained in Step 3 to the grinding system, collect actual feedback, calculate the deviation value using the large model, update the candidate solutions based on the deviation value, and return to Step 2 to re-evaluate the fitness of all candidate solutions until the grinding task is completed.

[0011] Furthermore, the parameters of the cuckoo algorithm include: population size, maximum number of iterations, basic elimination probability, capacity of the historical memory pool, local perturbation coefficient, initial step size, step size adaptive factor, elimination probability adjustment coefficient, and Levy distribution parameters.

[0012] Through the above operations, combined with the global search capability of the Cuckoo Optimization Algorithm and mechanisms such as adaptive step size, dynamic elimination, and local search, it can efficiently escape local optima in high-dimensional, multi-constraint parameter spaces, quickly obtain the optimal parameter combination, and improve the optimization speed and result stability.

[0013] Furthermore, the control parameters of the grinding system include feed rate, ball addition rate, water supply rate, and discharge concentration; The generation of candidate solutions based on the control parameters of the grinding system includes: randomly generating a predetermined number of candidate solutions within the preset physical boundaries of the control parameters of the grinding system, combined with a historical memory pool.

[0014] Furthermore, the addition of a global perturbation to all candidate solutions using the Levy flight mechanism is expressed by the following formula: ; in, For the first The generation A new solution for each candidate solution. For the first The generation There are 10 candidate solutions; For the first Motorcycle driver, Let be a Levy-distributed random variable, satisfying .

[0015] Furthermore, the large model employs an adaptive compensation adjustment mechanism to adjust the step size and improve convergence efficiency; The adaptive compensation adjustment mechanism is expressed by the following formula: ; in, Let this be the initial value of the step size. The difference between the optimal fitness scores of two adjacent generations. This is the highest rating in history. This is the step size adaptive factor.

[0016] Furthermore, the elimination mechanism based on fitness score includes: comparing the fitness score of the candidate solution with global perturbation with that of the corresponding candidate solution; if the fitness score of the candidate solution with global perturbation is higher than that of the corresponding candidate solution, the corresponding candidate solution is eliminated according to a predetermined elimination probability; otherwise, the candidate solution with global perturbation is removed.

[0017] Furthermore, during the elimination mechanism process, the predetermined elimination probability is adjusted based on fitness fluctuations, expressed by the following formula: ; in, To determine the probability of elimination, This is the elimination probability adjustment coefficient. For the first Standard deviation of generation fitness score For the first The optimal score is determined.

[0018] Furthermore, step 3 also includes perturbation processing of the candidate solution corresponding to the optimal fitness score to generate a local solution. If the fitness score of the local solution is higher than that of the candidate solution, the local solution replaces the candidate solution. The perturbation processing is expressed by the following formula: ; in, Indicates a local solution. Indicates the first The candidate solution corresponding to the optimal fitness. With a mean of 0 and a variance of The normal distribution perturbation.

[0019] Furthermore, during the execution of steps 3 and 4, if the grinding system control parameters corresponding to any candidate solution exceed the physical boundary, the projection method is used to pull back the physical boundary. The projection method is expressed by the following formula: ; in, For the first The solution is the first Dimensional parameters in the first dimension The value of the algebra, and This represents the physical upper and lower bounds of the parameter.

[0020] Through the above operations, the robustness of the grinding system is improved by using out-of-bounds projection, which can effectively cope with abnormal situations such as model errors, sensor failures, and extreme working conditions, and ensure the reliability and safety of the grinding system operation.

[0021] Furthermore, the step of collecting actual feedback, calculating deviation values ​​using a large model, and updating candidate solutions based on deviation values ​​includes: collecting grinding fineness and grinding efficiency under actual working conditions of the grinding system; performing actual scoring based on grinding fineness and grinding efficiency using predetermined rules; calculating deviation values ​​based on actual scores and fitness scores corresponding to the candidate solutions; and updating candidate solutions based on deviation values ​​using preset feedback calibration weights and uniform disturbance terms. The actual scoring based on the combination of grinding fineness and grinding efficiency according to predetermined rules includes: if the collected grinding fineness reaches the predetermined fineness index and the grinding efficiency increases, the actual score output is +100 points; if the collected grinding fineness reaches the predetermined fineness index and the grinding efficiency decreases, the actual score output is +50 points; if the collected grinding fineness reaches the predetermined fineness index and the grinding efficiency remains unchanged, the actual score output is 0 points; if the collected grinding fineness does not reach the predetermined fineness index and the grinding efficiency increases, the actual score output is -50 points; if the collected grinding fineness does not reach the predetermined fineness index and the grinding efficiency decreases, the actual score output is -100 points. The process of updating the candidate solution based on the deviation value, combined with preset feedback calibration weights and uniform perturbation terms, is expressed by the following formula: ; in, Indicates the updated candidate solution; Indicates the current optimal candidate solution; The feedback calibration weight is a preset value, specifically set to 0.2; To provide feedback on the difference between the model score and the actual score, It is a Gaussian or uniform perturbation term.

[0022] Through the above operations, a closed-loop optimization mechanism combining virtual model evaluation and actual process feedback is established. By dynamically calibrating the actual production feedback, the parameter optimization scheme can achieve continuous self-learning and adaptive adjustment, significantly enhancing the system's adaptability to real-world disturbances such as changes in operating conditions and equipment aging.

[0023] Beneficial effects: This invention provides a grinding system control method based on a large model and the Cuckoo algorithm, which realizes a closed-loop management of the entire process of large model evaluation, intelligent optimization and process parameter execution. The optimization results can directly guide actual production operations and improve the intelligence, automation and greenness of the grinding system.

[0024] It does not rely on specific equipment or hardware environment, has flexible parameter settings, is suitable for different types and scales of grinding systems, and has good engineering feasibility and industry promotion prospects. Attached Figure Description

[0025] Figure 1 This is a flowchart of a grinding system control method based on a large model and the cuckoo algorithm, according to an embodiment of the present invention. Detailed Implementation

[0026] The technical solution of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms "an" or "a" and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms "connected" or "linked" and similar terms are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up," "down," "left," "right," etc., are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship also changes accordingly.

[0028] Please see Figure 1 This application provides a method for controlling a grinding system based on a large model and the cuckoo algorithm, comprising: Step 1: Set the parameters of the Cuckoo Algorithm and generate candidate solutions based on the control parameters of the grinding system; In high-dimensional, multi-constrained, and dynamically changing grinding systems, existing optimization algorithms are prone to getting trapped in local optima, resulting in low optimization efficiency and difficulty in quickly obtaining the globally optimal parameter combination. This invention employs the Cuckoo Optimization Algorithm, combined with adaptive step size, dynamic elimination, and local search mechanisms, to significantly improve the global optimization capability and convergence speed of multi-parameter joint optimization.

[0029] The parameters of the Cuckoo algorithm include: population size, maximum number of iterations, basic elimination probability, capacity of the history memory pool, local perturbation coefficient, initial step size, step size adaptive factor, elimination probability adjustment coefficient, and Levy distribution parameters.

[0030] In this embodiment, the population size is set to 20, the maximum number of iterations is 50, the basic elimination probability is 0.2, the capacity of the history memory pool is set to 5, the local perturbation coefficient is set to 0.01, the initial step size is set to 0.5, the step size adaptive factor is set to 2, the elimination probability adjustment coefficient is set to 0.5, and the Levy distribution parameter is set to 1.5.

[0031] The control parameters of the grinding system include feed rate, ball rate, water feed rate, and discharge concentration; Generating candidate solutions based on grinding system control parameters includes: randomly generating a predetermined number of candidate solutions within the preset physical boundaries of the grinding system control parameters, combined with a historical memory pool constructed from historical grinding system control parameters; The generated candidate solution is a set of grinding system control parameters, which can be defined as: ; in, This represents one of the generated candidate solutions; This indicates the feed amount in the candidate solution set; This indicates the number of balls added in the candidate solutions. This indicates the water supply volume in the candidate solutions. This indicates the ore discharge concentration in the candidate solutions.

[0032] In this embodiment, the physical boundaries set for the above control parameters are as follows: Ore feed rate is greater than 80 tons / hour and less than 12080 tons / hour; ball addition rate is greater than 2 tons / hour and less than 5 tons / hour; water supply rate is greater than 30 cubic meters / hour and less than 50 cubic meters / hour; ore discharge concentration is greater than 65% and less than 75%.

[0033] Step 2: Use the large model to score the fitness of all candidate solutions and record the fitness of each candidate solution; Existing grinding systems are affected by various process parameters, and complex nonlinear coupling relationships exist between variables. Traditional mechanistic models and shallow data-driven models are insufficient to accurately characterize the dynamic behavior of the system, resulting in limited optimization and adjustment effects. This invention introduces an industrial-grade large-scale model to achieve high-precision evaluation and modeling of the efficiency of grinding systems under multiple parameters and complex operating conditions, providing a scientific and reliable objective function basis for intelligent optimization. Among them, the large model adopts an adaptive compensation adjustment mechanism to adjust the step size and improve convergence efficiency; The adaptive compensation adjustment mechanism is represented by the following formula: ; in, Let this be the initial value of the step size. The difference between the optimal fitness scores of two adjacent generations. This is the highest rating in history. This is the step size adaptive factor.

[0034] Step 3: Combine the Levy flight mechanism to add global perturbation to all candidate solutions, re-execute step 2, and perform the elimination mechanism based on fitness score. Repeat step 2 until the preset number of times is reached, and record the candidate solution corresponding to the best fitness score in the two fitness scoring processes. Incorporating the Levy flight mechanism, a global perturbation is added to all candidate solutions, expressed by the following formula: ; in, For the first The generation A new solution for each candidate solution. For the first The generation There are 10 candidate solutions; For the first Motorcycle driver, Let be a Levy-distributed random variable, satisfying , for Distribution parameters.

[0035] Specifically, the fitness scores of the candidate solutions with global perturbation are compared with those of the corresponding candidate solutions. If the fitness score of the candidate solution with global perturbation is higher than that of the corresponding candidate solution, the corresponding candidate solution is eliminated according to a predetermined elimination probability; otherwise, the candidate solution with global perturbation is removed.

[0036] During the elimination mechanism, the predetermined elimination probability is adjusted based on fitness fluctuations, as expressed by the following formula: ; in, To determine the probability of elimination, This is the elimination probability adjustment coefficient. For the first Standard deviation of generation fitness score For the first The optimal score is determined.

[0037] After selecting the candidate solution corresponding to the optimal fitness score, the candidate solution corresponding to the optimal fitness score is further perturbed to generate a local solution. If the fitness score of the local solution is higher than that of the candidate solution, the local solution is directly used to replace the candidate solution. Perturbation processing is expressed by the following formula: ; in, Indicates a local solution. Indicates the first The candidate solution corresponding to the optimal fitness. With a mean of 0 and a variance of The normal distribution perturbation.

[0038] Step 4: Apply the candidate solutions obtained in Step 3 to the grinding system, collect actual feedback, calculate the deviation value using the large model, update the candidate solutions based on the deviation value, and return to Step 2 to re-evaluate the fitness of all candidate solutions until the grinding task is completed.

[0039] During the execution of steps 3 and 4, if the control parameters of the grinding system corresponding to any candidate solution exceed the physical boundary, the projection method is used to pull back the physical boundary. The projection method is expressed by the following formula: ; in, For the first The solution is the first Dimensional parameters in the first dimension The value of the algebra, and This represents the physical upper and lower bounds of the parameter.

[0040] The process of collecting actual feedback, calculating deviation values ​​using a large model, and updating candidate solutions based on deviation values ​​includes: collecting grinding fineness and grinding efficiency under actual working conditions of the grinding system; performing actual scoring based on grinding fineness and grinding efficiency using predetermined rules; calculating deviation values ​​based on actual scores and fitness scores corresponding to the candidate solutions; and updating candidate solutions based on deviation values ​​using preset feedback calibration weights and uniform disturbance terms. The actual score is based on a combination of grinding fineness and grinding efficiency according to predetermined rules. The scores are as follows: if the collected grinding fineness reaches the predetermined fineness index and the grinding efficiency increases, the actual score is +100 points; if the collected grinding fineness reaches the predetermined fineness index and the grinding efficiency decreases, the actual score is +50 points; if the collected grinding fineness reaches the predetermined fineness index and the grinding efficiency remains unchanged, the actual score is 0 points; if the collected grinding fineness does not reach the predetermined fineness index and the grinding efficiency increases, the actual score is -50 points; if the collected grinding fineness does not reach the predetermined fineness index and the grinding efficiency decreases, the actual score is -100 points. The candidate solution is updated based on the deviation value, combined with the preset feedback calibration weights and uniform perturbation term, as expressed by the following formula: ; in, Indicates the updated candidate solution; Indicates the current optimal candidate solution; To provide feedback for calibration weights, To provide feedback on the difference between the model score and the actual score, In this embodiment, the disturbance term is either Gaussian or uniform. Set to 0.2, It follows a normal distribution with a mean of 0 and a standard deviation of 0.01, i.e. .

[0041] Currently, most grinding system optimization and control processes are open-loop or semi-closed-loop, lacking dynamic adaptive capabilities to actual operational feedback and effective compensation and fault-tolerance mechanisms for anomaly scoring and extreme operating conditions. This invention establishes a virtual-real combined closed-loop self-learning optimization framework, utilizing actual production feedback to continuously calibrate and optimize parameters, achieving robust adjustment to anomalies and disturbances, and ensuring long-term stable and efficient system operation.

[0042] Finally, in this embodiment, using the above parameter settings, a comparative experiment was conducted combining two typical optimization methods: large model with particle swarm optimization (PSO) and large model with cuckoo algorithm (CSA). The final fitness score, actual grinding efficiency improvement rate, and fitness fluctuation range were selected as evaluation indicators. The specific results are summarized in Table 1.

[0043] Table 1: Summary of comparative experimental results;

[0044] As can be seen from the comparative experimental results in Table 1, the method of this invention outperforms the large model combined with particle swarm optimization and the large model combined with traditional cuckoo algorithm in three key indicators: final fitness score, actual grinding efficiency improvement rate, and fitness fluctuation range. Specifically, the method of this invention effectively improves the accuracy and convergence speed of parameter optimization by introducing adaptive step size adjustment, dynamic elimination mechanism, and closed-loop calibration based on actual feedback, achieving a final fitness score of 95, significantly higher than the 83 and 91 scores of the comparative methods. Simultaneously, in terms of actual grinding efficiency improvement, this method achieves a 7% improvement, significantly higher than the traditional method, verifying the practical application value of the optimization results. Furthermore, the method of this invention has the lowest fitness fluctuation range, only 5%, demonstrating strong robustness and tolerance to disturbances, which helps the grinding system maintain stable operation under complex and variable working conditions. In summary, the experimental results fully demonstrate the significant advantages of the method of this invention in improving optimization effect and system stability.

[0045] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A control method for a grinding system based on a large model and the cuckoo algorithm, characterized in that, Includes the following steps: Step 1: Set the parameters of the Cuckoo Algorithm and generate candidate solutions based on the control parameters of the grinding system; Step 2: Use the large model to score the fitness of all candidate solutions and record the fitness of each candidate solution; Step 3: Combine the Levy flight mechanism to add global perturbation to all candidate solutions, re-execute step 2, and perform the elimination mechanism based on fitness score. Repeat step 2 until the preset number of times is reached, and record the candidate solution corresponding to the best fitness score in the two fitness scoring processes. Step 4: Apply the candidate solutions obtained in Step 3 to the grinding system, collect actual feedback, calculate the deviation value using the large model, update the candidate solutions based on the deviation value, and return to Step 2 to re-evaluate the fitness of all candidate solutions until the grinding task is completed.

2. The grinding system control method based on large model and cuckoo algorithm according to claim 1, characterized in that, The parameters of the Cuckoo Algorithm include: population size, maximum number of iterations, basic elimination probability, capacity of the historical memory pool, local perturbation coefficient, initial step size, step size adaptive factor, elimination probability adjustment coefficient, and Levy distribution parameters.

3. The grinding system control method based on large model and cuckoo algorithm according to claim 1, characterized in that, The control parameters of the grinding system include feed rate, ball addition rate, water supply rate, and discharge concentration; The generation of candidate solutions based on grinding system control parameters includes: randomly generating a predetermined number of candidate solutions within the preset physical boundaries of the grinding system control parameters, combined with a historical memory pool constructed from historical grinding system control parameters.

4. The grinding system control method based on large model and cuckoo algorithm according to claim 1, characterized in that, The addition of a global perturbation to all candidate solutions using the Levy flight mechanism is expressed by the following formula: ； in, For the first The generation A new solution for each candidate solution. For the first The generation There are 10 candidate solutions; For the first Motorcycle driver, Let be a Levy-distributed random variable, satisfying .

5. The grinding system control method based on large model and cuckoo algorithm according to claim 4, characterized in that, The large model employs an adaptive compensation adjustment mechanism to adjust the step size and improve convergence efficiency. The adaptive compensation adjustment mechanism is expressed by the following formula: ； in, Let this be the initial value of the step size. The difference between the optimal fitness scores of two adjacent generations. This is the highest rating in history. This is the step size adaptive factor.

6. The grinding system control method based on large model and cuckoo algorithm according to claim 1, characterized in that, The elimination mechanism based on fitness score includes: comparing the fitness score of the candidate solution with global perturbation with that of the corresponding candidate solution; if the fitness score of the candidate solution with global perturbation is higher than that of the corresponding candidate solution, the corresponding candidate solution is eliminated according to a predetermined elimination probability; otherwise, the candidate solution with global perturbation is removed.

7. The grinding system control method based on large model and cuckoo algorithm according to claim 6, characterized in that, During the elimination mechanism, the predetermined elimination probability is adjusted based on fitness fluctuations, as expressed by the following formula: ； in, To determine the probability of elimination, This is the elimination probability adjustment coefficient. For the first Standard deviation of generation fitness score For the first The optimal score is determined.

8. The grinding system control method based on large model and cuckoo algorithm according to claim 1, characterized in that, Step 3 also includes perturbating the candidate solution corresponding to the optimal fitness score to generate a local solution. If the fitness score of the local solution is higher than that of the candidate solution, the local solution replaces the candidate solution. The perturbation processing is expressed by the following formula: ； in, Indicates a local solution. Indicates the first The candidate solution corresponding to the optimal fitness. With a mean of 0 and a variance of The normal distribution perturbation.

9. The grinding system control method based on large model and cuckoo algorithm according to claim 1, characterized in that, During the execution of steps 3 and 4, if the grinding system control parameters corresponding to any candidate solution exceed the physical boundary, the projection method is used to pull back the physical boundary. The projection method is expressed by the following formula: ； in, For the first The solution is the first Dimensional parameters in the first dimension The value of the algebra, and This represents the physical upper and lower bounds of the parameter.

10. The grinding system control method based on large model and cuckoo algorithm according to any one of claims 1-9, characterized in that, The process of collecting actual feedback, calculating deviation values ​​using a large model, and updating candidate solutions based on deviation values ​​includes: collecting grinding fineness and grinding efficiency under actual working conditions of the grinding system; performing actual scoring based on grinding fineness and grinding efficiency using predetermined rules; calculating deviation values ​​based on actual scores and fitness scores corresponding to the candidate solutions; and updating candidate solutions based on deviation values ​​using preset feedback calibration weights and uniform disturbance terms. The actual scoring based on the combination of grinding fineness and grinding efficiency according to predetermined rules includes: if the collected grinding fineness reaches the predetermined fineness index and the grinding efficiency increases, the actual score output is +100 points; if the collected grinding fineness reaches the predetermined fineness index and the grinding efficiency decreases, the actual score output is +50 points; if the collected grinding fineness reaches the predetermined fineness index and the grinding efficiency remains unchanged, the actual score output is 0 points; if the collected grinding fineness does not reach the predetermined fineness index and the grinding efficiency increases, the actual score output is -50 points; if the collected grinding fineness does not reach the predetermined fineness index and the grinding efficiency decreases, the actual score output is -100 points. The process of updating the candidate solution based on the deviation value, combined with preset feedback calibration weights and uniform perturbation terms, is expressed by the following formula: ； in, Indicates the updated candidate solution; Indicates the current optimal candidate solution; To provide feedback for calibration weights, To provide feedback on the difference between the model score and the actual score, It is a Gaussian or uniform perturbation term.

Images