Fuzzy and reinforced learning collaborative graphitization furnace regulation and control method and system

Abstract

The invention discloses a fuzzy and reinforcement learning collaborative graphitization furnace regulation and control method and system. The regulation and control method comprises the following steps: continuously obtaining working condition parameters; calculating a deviation value between the actual value of the working condition parameter and a preset standard value at regular time, and judging whether the deviation value meets a preset abnormal condition or not so as to select a first method / a second method for regulation and control; the first method comprises the following steps: performing fuzzy reasoning based on the deviation value to obtain a first adjusting signal to adjust the graphitization furnace; the second method comprises the following steps: processing the current working condition parameters of the graphitization furnace through the reinforcement learning model to obtain a second adjusting signal to adjust the graphitization furnace. According to the graphitization furnace regulation and control method based on the fuzzy and reinforcement learning cooperation, the stability of operation of the machining program of the graphitization furnace can be enhanced, and the production efficiency of products is improved.

Description

Technical Field

[0001] This invention relates to the field of graphitization furnace operation control technology, and in particular to a graphitization furnace control method and system based on fuzzy and reinforcement learning collaboration. Background Technology

[0002] Graphitization technology, as a core process in carbon material production, holds an irreplaceable strategic position in fields such as lithium-ion battery anode materials, graphite electrodes, and specialty carbon products. Traditional intermittent graphitization processes have long dominated, relying on Atchison furnaces or internal series graphitization furnaces for single-furnace production. This model has significant technical drawbacks: single-furnace cycles are long (15-30 days), energy consumption is high, temperature uniformity is difficult to control precisely, and the physical properties of products fluctuate significantly between batches, making it difficult to meet the stringent requirements for material consistency in high-end applications.

[0003] To overcome the aforementioned limitations, continuous graphitization technology has emerged. This technology achieves continuous material feeding through high-temperature tubular or vertical furnaces, theoretically significantly reducing repetitive heating and cooling processes and the number of equipment start-ups and shutdowns, thereby substantially reducing energy consumption per unit product and production cycle time, demonstrating significant energy-saving potential. However, existing continuous graphitization processes still face two major bottlenecks in practical industrial applications: First, the dynamic matching of key process parameters such as temperature field distribution, material residence time, and atmosphere control is difficult, and actual production capacity is still greatly constrained. Second, the real-time control capability of operating parameters (such as heating temperature) is insufficient, and the degree of adjustment cannot be adjusted according to actual changes, which easily leads to the product qualification rate being lower than expected for a long time, thus restricting the promotion and production of continuous graphitization.

[0004] To address the aforementioned technical challenges, this invention proposes an intelligent control system based on an adaptive algorithm. This system aims to simultaneously solve the technical problems of low production efficiency and poor operational stability in continuous graphitization processes by optimizing process parameters in real time, thus providing an innovative solution for the efficient and stable production of carbon materials. Summary of the Invention

[0005] The purpose of this invention is to provide a fuzzy and reinforcement learning collaborative graphitization furnace control method and system that can dynamically adjust the operating parameters based on the actual working conditions of the graphitization furnace to solve the problems of low production efficiency and poor operational stability in continuous graphitization processes.

[0006] To achieve the above objectives, this invention discloses a method for controlling a graphitization furnace using a combination of fuzzy and reinforcement learning, comprising: Continuously acquire operating parameters representing the working state of the heating zone in the graphitization furnace heating zone; The deviation between the actual value of the operating condition parameter and the preset standard value is calculated periodically, and it is determined whether the deviation meets the preset abnormal conditions. If yes, the adjustment is made according to the first method below; if no, the adjustment is made according to the second method below. The first method is as follows: performing fuzzy inference based on the deviation value to obtain a first adjustment parameter, and performing calculation based on the first adjustment parameter and the deviation value to obtain a first adjustment signal; The graphitization furnace is adjusted based on the first adjustment signal; The second method is: to construct a reinforcement learning model based on the operating parameters of the graphitization furnace under different working scenarios; The reinforcement learning model processes the current operating parameters of the graphitization furnace to obtain a second adjustment parameter, and calculates a second adjustment signal based on the deviation between the actual value of the second adjustment parameter and the operating parameters and the preset standard value. The graphitization furnace is adjusted based on the second adjustment signal.

[0007] Specifically, the first adjustment parameter includes a first proportional parameter, a first integral parameter, and a first derivative parameter; the second adjustment parameter includes a second proportional parameter, a second integral parameter, and a second derivative parameter. The calculation formula for the first adjustment signal or the second adjustment signal is as follows:

[0008] in, The deviation between the actual value of the operating condition parameter and the preset standard value, when When the first adjustment signal is used, This is the first proportional parameter. The first integral parameter, This is the first differential parameter; when When it is the second adjustment signal, This is the second proportional parameter. The second integral parameter is... is the second differential parameter.

[0009] Furthermore, the fuzzy reasoning process includes: Based on the deviation between the actual value and the preset standard value of the operating condition parameter, the rate of change of the deviation value is calculated periodically. Calculate the deviation between the actual value of the current operating condition parameter and the preset standard value, and the first and second membership degrees of the current deviation change rate in the preset fuzzy set. Based on the first membership degree and the second membership degree, calculate the activation strength of each rule in the preset fuzzy rule base; The first adjustment parameter is obtained by defuzzifying the activation intensity of each rule.

[0010] Specifically, the method for constructing the reinforcement learning model includes: Based on the operating condition parameters, configure the state space, the action space representing the adjustment action of the second proportional parameter, the second integral parameter, and the second derivative parameter, and the expected value function; A neural network model is provided, and reinforcement learning training is performed on the neural network model based on the action space, the state space, and the expected value function to obtain the reinforcement learning model.

[0011] Furthermore, the state space includes environmental parameters representing different working environments of the graphitization furnace, and the reinforcement learning training steps include: Step 1: Initialize the expected values ​​of each adjustment action in the action space; Step 2: Preset a set of the second proportional parameters, the second integral parameters, and the second differential parameters; Step 3: Under different environmental parameter values, perform each adjustment action in the action space to obtain the changed operating condition parameters, the second proportional parameter, the second integral parameter, the second derivative parameter, and the environmental parameters after execution; Step 4: Based on the operating condition parameters and environmental parameters before and after each adjustment action in the action space, calculate the expected value corresponding to the currently executed adjustment action in the action space; Repeat steps three and four until the preset stop condition is met. The set of adjustment actions in the action space that have been executed with the highest expected value under each of the environmental parameters is taken as the optimal adjustment step. The corresponding second proportional parameter, second integral parameter and second derivative parameter obtained by adjusting based on the optimal adjustment step are taken as the optimal second control parameters to calculate the second adjustment signal.

[0012] Furthermore, the expected value is updated once for each adjustment action in the action space, and the expected value function is:

[0013] in, For the current environmental parameters The adjustment action in the action space is executed below. The calculated expected value, The environmental parameters that change after performing the adjustment action in the action space. The preset learning rate, The reward coefficient is... The preset discount factor, To perform the execution action The environmental parameters obtained afterwards Execution actions in the action space that can be executed below. The corresponding maximum value of the expected value.

[0014] The present invention also discloses a graphitization furnace control system based on fuzzy and reinforcement learning, which operates based on the fuzzy and reinforcement learning co-operated graphitization furnace control method described above.

[0015] This invention also discloses a fuzzy and reinforcement learning collaborative graphitization furnace control system, which includes: One or more processors; Memory; And one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs including instructions for performing the fuzzy and reinforcement learning synergistic graphitization furnace control method as described above.

[0016] The present invention also discloses a computer-readable storage medium, characterized in that it includes a computer program, which can be executed by a processor to perform the fuzzy and reinforcement learning collaborative graphitization furnace control method as described above.

[0017] Compared with existing technologies, the fuzzy and reinforcement learning-based graphitization furnace control method provided by the present invention achieves a dual improvement in process optimization and stable control by continuously monitoring the operating parameters of the heating zone and dynamically adjusting the control method. This control method first acquires operating parameters representing the working state of the heating zone in real time, and performs deviation analysis based on preset standard values, enabling rapid triggering of targeted adjustment mechanisms. When the deviation meets preset abnormal conditions, a first adjustment signal is generated using fuzzy inference. This algorithm has a fast response speed and can effectively handle complex operating conditions with strong nonlinearity and time-varying characteristics, ensuring rapid response and coarse adjustment stability under abnormal conditions. Within the normal deviation range, a second adjustment signal is generated by relying on a reinforcement learning model to continuously learn the patterns associated with operating parameters under different working scenarios, achieving fine-grained optimization and adaptive fine-tuning. This dual-mode switching mechanism ensures safe control under abnormal operating conditions and improves efficiency and accuracy during steady-state operation, reducing energy waste and product performance fluctuations caused by fixed parameter control in traditional methods. Compared to existing technologies, this solution upgrades process parameter control from passive response to active prediction. Fuzzy reasoning and reinforcement learning methods enhance the system's adaptability to complex operating conditions, thereby improving the utilization rate of graphitization furnace capacity while effectively improving the operational stability of the graphitization furnace and the consistency of product quality. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the graphitization furnace control method using fuzzy and reinforcement learning in an embodiment of the present invention.

[0019] Figure 2 This is a schematic diagram of the fuzzy reasoning process in an embodiment of the present invention.

[0020] Figure 3 This is a schematic diagram of the construction method of the reinforcement learning model in an embodiment of the present invention.

[0021] Figure 4 This is a schematic diagram of the reinforcement learning training process in an embodiment of the present invention. Detailed Implementation

[0022] To illustrate the technical content, structural features, objectives, and effects of the present invention in detail, the following description is provided in conjunction with the embodiments and accompanying drawings.

[0023] This invention discloses a fuzzy and reinforcement learning collaborative graphitization furnace control method to improve the problems of low production efficiency and poor operational stability in continuous graphitization processes. (See also...) Figure 1 As shown, the fuzzy and reinforcement learning collaborative graphitization furnace control method specifically includes the following steps: S1: Continuously acquire operating parameters representing the working state of the heating zone in the graphitization furnace. In this embodiment, the operating parameters include heating temperature, gas pressure within the heating zone, flow rate of the material being heated in the heating zone, and composition of the protective gas, etc.

[0024] S2: Periodically calculate the deviation between the actual value of the operating condition parameter and the preset standard value. In this embodiment, the deviation value is the difference between the actual value of the operating condition parameter and the preset standard value. In this embodiment, the preset standard value is determined by factors such as the actual scale and production horsepower of the graphitization furnace. Assuming the operating condition parameter is the heating temperature of the heating zone, the preset standard value is the target temperature that the heating zone should currently achieve.

[0025] S3: Determine whether the deviation between the actual value of the operating condition parameter and the preset standard value meets the preset abnormal conditions. If yes, proceed to step S4; otherwise, proceed to step S6.

[0026] S4: Perform fuzzy inference based on the deviation between the actual value and the preset standard value of the operating condition parameter to obtain the first adjustment parameter. Calculate based on the first adjustment parameter and the deviation value to obtain the first adjustment signal, and proceed to step S5.

[0027] S5: Adjust the graphitization furnace based on the first adjustment signal and return to step S2.

[0028] S6: Based on the operating parameters of the graphitization furnace under different working scenarios, construct a reinforcement learning model and proceed to step S7.

[0029] S7: The current operating parameters of the graphitization furnace are processed by a reinforcement learning model to obtain a second adjustment parameter. The second adjustment signal is obtained by calculating the deviation between the actual value of the second adjustment parameter and the operating parameter and the preset standard value.

[0030] S8: Adjust the graphitization furnace based on the second adjustment signal.

[0031] Compared with existing technologies, the graphitization furnace control method proposed in this invention, which combines fuzzy and reinforcement learning, realizes the transformation of process control from experience-dependent to data-driven by constructing a dynamic parameter monitoring and intelligent adjustment system. The control method uses the operating parameters of the heating zone as the monitoring object and analyzes the graphitization furnace in conjunction with preset standard values.

[0032] When a deviation value is detected that meets a preset abnormal condition (in this embodiment, i.e., when the deviation value exceeds a safety threshold), the control method immediately activates the first method to perform fuzzy inference. Fuzzy inference, based on a preset set of fuzzy rules, can quickly transform complex nonlinear and time-varying operating conditions into quantifiable adjustment commands, thereby generating a first adjustment signal. This ensures rapid response and coarse-adjustment stability of the graphitization furnace control under abnormal conditions, effectively avoiding equipment damage or product quality defects caused by parameter runaway. During the steady-state operation phase where the deviation value does not meet the abnormal conditions, the system automatically switches to reinforcement learning mode. By constructing a reinforcement learning model, it continuously learns the dynamic correlation patterns of operating parameters under different production scenarios of the graphitization furnace, thereby generating a second adjustment signal. The reinforcement learning adjustment method is relatively slow and stable, achieving adaptive fine-tuning.

[0033] The advantage of this dual-mode collaborative control mechanism lies in its integration of fuzzy logic and reinforcement learning. Under abnormal operating conditions, the rapid response characteristics of fuzzy inference can effectively shorten the adjustment delay, bringing the graphitization furnace back to a more stable operating state. In steady-state operation, the reinforcement learning model can dynamically compensate for the adjustment lag caused by fixed parameters in traditional control. The stable adjustment of the reinforcement learning model can reduce energy consumption fluctuations and product performance dispersion.

[0034] In summary, this invention enhances the adaptability of graphitization furnaces to different working environments through the synergistic effect of fuzzy reasoning and reinforcement learning, and improves the problems of low efficiency and poor stability in traditional graphitization processes.

[0035] Specifically, the first adjustment parameter includes a first proportional parameter, a first integral parameter, and a first derivative parameter; the second adjustment parameter includes a second proportional parameter, a second integral parameter, and a second derivative parameter.

[0036] The formula for calculating the first or second adjustment signal is:

[0037] in, The deviation between the actual value and the preset standard value of the operating condition parameter, when When it is the first adjustment signal, The first proportional parameter, The first integration parameter, The first differential parameter; when When it is the second adjustment signal, This is the second proportional parameter. For the second integral parameter, is the second differential parameter.

[0038] Specifically, see Figure 2As shown, the process of fuzzy reasoning includes: S41: Based on the deviation between the actual value and the preset standard value of the operating condition parameter, calculate the rate of change of the deviation value periodically. In this embodiment, the rate of change of the deviation value is the rate of change of the deviation value with respect to time, i.e. .

[0039] S42: Calculate the deviation between the actual value and the preset standard value of the current operating condition parameter, and the first and second membership degrees of the current deviation change rate to the preset fuzzy set. Fuzzy sets are mathematical tools for describing fuzzy concepts, allowing elements to belong to a set with different degrees of membership. The membership degree of an element can be calculated using a membership function. Common membership functions include trigonometric functions, trapezoidal functions, Gaussian functions, sigmoid functions, and generalized Bell functions.

[0040] S43: Based on the first membership degree and the second membership degree, calculate the activation strength of each rule in the preset fuzzy rule base.

[0041] S44: Defuzzify based on the activation intensity of each rule to obtain the first adjustment parameter.

[0042] In the process of fuzzy inference, the fuzzy set can be set to seven subsets, namely "positive large, positive medium, positive small, zero, negative small, negative medium, and negative large", depending on the actual situation. Taking a fuzzy rule as an example, the fuzzy rule base can be set in the following form: IF is “burden” AND is “burdened” THEN ↑30%, ↑20%, ↓10%.

[0043] Based on multiple preset fuzzy rules and deviation values and the rate of change of deviation value The first and second membership degrees can then be obtained. Based on these membership degrees, the activation strength of each rule in the fuzzy rule base can be calculated. Based on the activation strength of each rule, defuzzification can be performed using any of the following methods: centroid method, maximum rule activation method, or weighted average method, to obtain the final first adjustment parameter. The specific process of fuzzy inference is understood by those skilled in the art, and the principles will not be elaborated upon here.

[0044] Specifically, see Figure 3 As shown, the methods for constructing reinforcement learning models include: S61: Configure the state space based on the operating condition parameters, the action space representing the adjustment actions of the second proportional parameter, the second integral parameter, and the second derivative parameter, and the expected value function. In this embodiment, the adjustment actions of the action space include the adjustment of the second proportional parameter, the second integral parameter, and the second derivative parameter, and the adjustment amount at one time is preferably [-0.2, +0.2].

[0045] S62: Provides a neural network model, and performs reinforcement learning training on the neural network model based on the action space, state space and expectation function to obtain a reinforcement learning model.

[0046] Furthermore, the state space includes environmental parameters representing different working environments of the graphitization furnace, see [reference]. Figure 4 As shown, in this embodiment, reinforcement learning is trained based on the Q-Learning algorithm. The steps of reinforcement learning training include: S621: Initialize the expected values ​​of each adjustment action in the action space.

[0047] S622: Preset a set of second proportional parameters, second integral parameters and second differential parameters.

[0048] S623: Under different environmental parameter values, execute various adjustment actions in the action space to obtain the changed operating parameters, second proportional parameter, second integral parameter, second derivative parameter, and environmental parameters after execution. In this embodiment, it is assumed that the monitored operating parameter is the heating temperature, and the environmental parameters include the flow rate, pressure, and feed rate of the protective gas, etc. If the monitored operating parameter is the feed rate, then the corresponding environmental parameters may be the heating temperature, the flow rate of the protective gas, etc.

[0049] S624: Based on the operating condition parameters and environmental parameters before and after each adjustment action in the action space, calculate the expected value corresponding to the adjustment action in the currently executed action space.

[0050] S625: Determine whether the preset stopping condition is met. If yes, proceed to step S626; otherwise, return to step S623. The preset stopping condition is preferably the number of rounds of reinforcement learning training. When the number of rounds of reinforcement learning training reaches a preset value, the reinforcement learning training stops.

[0051] S626: The set of adjustment actions in the executed action space with the highest expected value under each environmental parameter is taken as the optimal adjustment step, and the corresponding second proportional parameter, second integral parameter and second derivative parameter obtained based on the optimal adjustment step are taken as the optimal second control parameters to calculate the second adjustment signal.

[0052] In this embodiment, the reinforcement learning model will continuously undergo reinforcement learning training so that the optimal adjustment steps will be continuously updated and iterated, thereby making the corresponding second proportional parameters, second integral parameters, and second differential parameters obtained through the optimal adjustment steps more adaptable to the working process of the graphitization furnace.

[0053] Furthermore, the expected value is updated once for each adjustment action in the action space, and the expected value function is:

[0054] in, In the current environmental parameters Adjustment action in the lower execution action space The calculated expected value, To adjust the environmental parameters that change after the action in the action space. The preset learning rate, As the reward coefficient, The preset discount factor, To perform the execution action The environmental parameters obtained later Execution of actions in all possible action spaces The corresponding maximum expected value. In this embodiment, the learning rate... The preferred range is [0.01, 2], discount factor The preferred range is [0.9, 0.99], and the reward coefficient should be set in combination with the operating parameters and environmental parameters.

[0055] Still assuming the monitored operating parameters are heating temperature and the environmental parameters are feed rate, protective gas flow rate, and energy consumption, the reward coefficient function can be set as follows:

[0056] in, As the reward coefficient, For heating temperature tracking, For feed stability, For airflow optimization, This is an energy consumption penalty item. Temperature weighting, For feed rate weighting, For traffic weight and As an energy consumption weight, preferably, , ＞ ＞ ＞ , This represents the deviation between the actual heating temperature and the preset standard value. This is the preset standard value for the heating temperature. The change in feed rate The baseline feed rate is (e.g., 300 mm / min). For the ideal flow rate of the heating process, For real-time traffic, The baseline flow rate (often taken as the baseline flow rate) ), Power consumption base, power consumption base = Energy consumption per unit time / baseline energy consumption, where the baseline energy consumption is the minimum energy consumption under ideal operating conditions.

[0057] The present invention also discloses a graphitization furnace control system based on fuzzy and reinforcement learning, which operates based on the fuzzy and reinforcement learning-based graphitization furnace control method described above.

[0058] This invention also discloses another fuzzy and reinforcement learning collaborative graphitization furnace control system, which includes one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors. The programs include instructions for performing the fuzzy and reinforcement learning collaborative graphitization furnace control method as described above. The processors can be general-purpose central processing units (CPUs), microprocessors, application-specific integrated circuits (ASICs), or one or more integrated circuits, used to execute relevant programs to implement the functions required by the modules in the fuzzy and reinforcement learning collaborative graphitization furnace control system of this application, or to execute the fuzzy and reinforcement learning collaborative graphitization furnace control method of this application.

[0059] This invention also discloses a computer-readable storage medium comprising a computer program executable by a processor to perform the graphitization furnace control method for fuzzy and reinforcement learning collaboration as described above. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center integrating one or more available media. The available medium can be read-only memory (ROM), random access memory (RAM), or magnetic media, such as floppy disks, hard disks, magnetic tapes, magnetic disks, or optical media, such as digital versatile discs (DVDs), or semiconductor media, such as solid-state disks (SSDs).

[0060] The above-disclosed embodiments are merely preferred embodiments of the present invention and should not be construed as limiting the scope of the present invention. Therefore, any equivalent variations made in accordance with the claims of the present invention are still within the scope of the present invention.

Claims

1. A method for controlling a graphitization furnace using a combination of fuzzy and reinforcement learning, characterized in that, include: Continuously acquire operating parameters representing the working state of the heating zone in the graphitization furnace heating zone; The deviation between the actual value of the operating condition parameter and the preset standard value is calculated periodically, and it is determined whether the deviation meets the preset abnormal conditions. If yes, the adjustment is made according to the first method below; if no, the adjustment is made according to the second method below. The first method is as follows: performing fuzzy inference based on the deviation value to obtain a first adjustment parameter, and performing calculation based on the first adjustment parameter and the deviation value to obtain a first adjustment signal; The graphitization furnace is adjusted based on the first adjustment signal; The second method is: to construct a reinforcement learning model based on the operating parameters of the graphitization furnace under different working scenarios; The reinforcement learning model processes the current operating parameters of the graphitization furnace to obtain a second adjustment parameter, and calculates a second adjustment signal based on the deviation between the actual value of the second adjustment parameter and the operating parameters and the preset standard value. The graphitization furnace is adjusted based on the second adjustment signal.

2. The method for controlling a graphitization furnace using fuzzy and reinforcement learning synergy according to claim 1, characterized in that, The first adjustment parameter includes a first proportional parameter, a first integral parameter, and a first derivative parameter; the second adjustment parameter includes a second proportional parameter, a second integral parameter, and a second derivative parameter. The calculation formula for the first adjustment signal or the second adjustment signal is as follows: in, The deviation between the actual value of the operating condition parameter and the preset standard value, when When the first adjustment signal is used, This is the first proportional parameter. The first integral parameter, This is the first differential parameter; when When it is the second adjustment signal, This is the second proportional parameter. The second integral parameter is... is the second differential parameter.

3. The method for controlling a graphitization furnace using fuzzy and reinforcement learning synergy according to claim 2, characterized in that, The fuzzy reasoning process includes: Based on the deviation between the actual value and the preset standard value of the operating condition parameter, the rate of change of the deviation value is calculated periodically. Calculate the deviation between the actual value of the current operating condition parameter and the preset standard value, and the first and second membership degrees of the current deviation change rate in the preset fuzzy set. Based on the first membership degree and the second membership degree, calculate the activation strength of each rule in the preset fuzzy rule base; The first adjustment parameter is obtained by defuzzifying the activation intensity of each rule.

4. The fuzzy and reinforcement learning collaborative graphitization furnace control method according to claim 2, characterized in that, The method for constructing the reinforcement learning model includes: Based on the operating condition parameters, configure the state space, the action space representing the adjustment action of the second proportional parameter, the second integral parameter, and the second derivative parameter, and the expected value function; A neural network model is provided, and reinforcement learning training is performed on the neural network model based on the action space, the state space, and the expected value function to obtain the reinforcement learning model.

5. The method for controlling a graphitization furnace using fuzzy and reinforcement learning synergy according to claim 4, characterized in that, The state space includes environmental parameters representing different working environments of the graphitization furnace, and the reinforcement learning training steps include: Step 1: Initialize the expected values ​​of each adjustment action in the action space; Step 2: Preset a set of the second proportional parameters, the second integral parameters, and the second differential parameters; Step 3: Under different environmental parameter values, perform each adjustment action in the action space to obtain the changed operating condition parameters, the second proportional parameter, the second integral parameter, the second derivative parameter, and the environmental parameters after execution; Step 4: Based on the operating condition parameters and environmental parameters before and after each adjustment action in the action space, calculate the expected value corresponding to the currently executed adjustment action in the action space; Repeat steps three and four until the preset stop condition is met. The set of adjustment actions in the action space that have been executed with the highest expected value under each of the environmental parameters is taken as the optimal adjustment step. The corresponding second proportional parameter, second integral parameter and second derivative parameter obtained by adjusting based on the optimal adjustment step are taken as the optimal second control parameters to calculate the second adjustment signal.

6. The method for controlling a graphitization furnace using fuzzy and reinforcement learning collaboration according to claim 5, characterized in that, For each adjustment action performed in the aforementioned action space, the expected value is updated once, and the expected value function is: in, For the current environmental parameters The adjustment action in the action space is executed below. The calculated expected value, The environmental parameters that change after performing the adjustment action in the action space. The preset learning rate, The reward coefficient is... The preset discount factor, To perform the execution action The environmental parameters obtained afterwards Execution actions in the action space that can be executed below. The corresponding maximum value of the expected value.

7. A graphitization furnace control system based on fuzzy and reinforcement learning, characterized in that, The fuzzy and reinforcement learning collaborative graphitization furnace control system operates based on the fuzzy and reinforcement learning collaborative graphitization furnace control method according to any one of claims 1 to 6.

8. A graphitization furnace control system that combines fuzzy and reinforcement learning, characterized in that, include: One or more processors; Memory; And one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs including instructions for performing the fuzzy and reinforcement learning collaborative graphitization furnace control method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, Includes a computer program that can be executed by a processor to perform the fuzzy and reinforcement learning collaborative graphitization furnace control method as described in any one of claims 1 to 6.

Images