Reinforcement learning driven intelligent temperature control joule heating amplification device and method thereof

Abstract

The application provides a kind of reinforcement learning driven intelligent temperature control joule heating amplification device and method.The device includes scalable power modulation module, modular reaction cavity and multi-point electrode array structure, which can realize flexible expansion of reaction volume, electrode number and heating area, and is suitable for different preparation scale requirements.The application is based on COMSOL-Simulink joint simulation to build joule heat-electricity coupling model, and introduces reinforcement learning intelligent control method to complete control strategy training in simulation environment;During device operation, temperature feedback is obtained through multi-point temperature sensor, and intelligent controller adaptively adjusts pulse voltage amplitude, width and frequency to realize multi-point electrode collaborative heating and high-precision temperature control.This method does not require manual parameter setting, has good robustness and scalability, and can significantly improve temperature field uniformity, energy efficiency and large-scale preparation capacity.

Description

Technical Field

[0001] This invention relates to the field of materials preparation and heat treatment technology, specifically to a reinforcement learning-driven intelligent temperature control Joule heating amplification device and a multi-point electrode intelligent temperature control method for large-scale preparation. Background Technology

[0002] Flash Joule heating (FJH) is a novel rapid high-temperature processing technology that uses a high-energy pulsed current applied for an extremely short time to instantly raise the temperature of materials to thousands of degrees Celsius within hundreds of milliseconds to several seconds. This enables the graphitization of carbon materials, the reduction of metal oxides, and the structural control of composite materials. Due to its extremely fast heating rate and high energy efficiency, FJH technology has shown significant advantages and broad application prospects in fields such as carbon material preparation, resource recycling, and the development of new energy devices.

[0003] However, existing FJH devices mostly adopt a single-point electrode structure and fixed cavity design, which is only suitable for small-scale laboratory preparation and has several limitations: First, the reaction chamber volume is limited, making it difficult to meet the needs of large-scale preparation at the gram level or above; second, the temperature distribution is uneven in the single-point electrode mode, and the internal temperature gradient of the sample is large, resulting in fluctuations in product performance; third, the device's scalability and adaptability are insufficient, making it difficult to meet the requirements of continuous preparation under different yields and different material systems.

[0004] Furthermore, in terms of temperature control strategies, traditional FJH devices are mostly based on constant voltage or constant current drive methods, which are typical single-variable control strategies. These methods typically adjust only a single control parameter (such as voltage amplitude). While simple in structure and fast in response, they have significant shortcomings in complex flash heating processes: First, single-variable control lacks the ability to coordinate parameters such as pulse width and frequency, failing to flexibly address the thermal response characteristics of materials at different stages, resulting in a single energy input mode; second, due to the strong nonlinearity, strong coupling, and time-varying nature of the Joule heating process, single voltage adjustment is insufficient to achieve precise and stable control of the temperature field, easily leading to local overheating or uneven temperature distribution; third, constant voltage drive struggles to simultaneously meet multiple objectives such as temperature accuracy, heating uniformity, and energy consumption optimization, affecting product structure consistency and energy utilization efficiency.

[0005] In summary, existing FJH devices still face significant bottlenecks in terms of preparation scale, temperature field uniformity, and intelligent control. Therefore, there is an urgent need to develop a novel Joule thermal flash evaporation device with a scalable reaction chamber structure, a multi-point electrode array heating module, and a multivariable intelligent temperature control strategy. This would enable dynamic and precise temperature field control, adaptive optimization of energy input, and promote the development of FJH technology towards large-scale, controllable, and industrialized applications. Summary of the Invention

[0006] The purpose of this invention is to solve the problems of limited preparation capacity, uneven temperature distribution, single control method, and poor system scalability in existing Joule thermal flash evaporation devices.

[0007] The objective of this invention can be achieved through the following technical solutions: A reinforcement learning-driven intelligent temperature-controlled Joule heating amplification device includes a power supply and control unit, an energy storage and pulse modulation unit, a reaction chamber unit, a multi-point electrode array unit, and a temperature monitoring and intelligent control unit. The power supply and control unit includes a constant voltage power supply module and a pulse power supply module, which are used to output a constant voltage or a pulse voltage signal with adjustable amplitude, width and frequency. The energy storage and pulse modulation unit includes a capacitor energy storage module, a power switch module, and a filter unit, used to modulate the amplitude, width, and frequency of the pulse voltage; The reaction chamber unit adopts a modular and expandable structure, and the chamber size can be enlarged or reduced according to the preparation requirements. The multi-point electrode array unit is distributed with multiple graphite or metal electrodes along the inside of the reaction chamber to achieve multi-point heating of the sample and reduce the temperature gradient. The temperature monitoring and intelligent control unit includes a multi-point temperature sensor, a PLC controller, and a reinforcement learning control module; the multi-point temperature sensor is used to collect temperature data of multiple heating points in the reaction chamber in real time. The reinforcement learning control module adaptively adjusts the input voltage of the multi-point electrodes based on the feedback signals from the multi-point temperature sensors, thereby achieving dynamic and coordinated control of the temperature field of the reaction chamber.

[0008] Preferably, the multi-point temperature sensor adopts a distributed thermocouple or infrared temperature sensor to collect temperature data of each heating point in the reaction chamber in real time, form multi-dimensional temperature status data, and transmit the data to the central control system. The reinforcement learning control module generates an optimized control signal to realize closed-loop temperature regulation.

[0009] Preferably, the reinforcement learning algorithm module adaptively optimizes the amplitude, width, and frequency of the pulse voltage based on the interaction between the environmental state, control action, and reward function, so as to achieve high-precision control of the temperature field.

[0010] Preferably, the reaction chamber unit is connected in series or in parallel with multiple reaction chamber units through a modular interface to expand the production capacity and meet the production needs of different scales.

[0011] Preferably, the number of electrodes, electrode spacing, electrode arrangement, and electrode material of the multi-point electrode array are adjusted according to the sample properties and heat treatment process requirements to optimize the uniformity of current and temperature distribution.

[0012] This invention also provides a method for a reinforcement learning-driven intelligent temperature-controlled Joule heating amplification device, comprising the following steps: Step S1: Collect temperature data inside the reaction chamber in real time using multi-point temperature sensors; Step S2: Input the temperature data into the reinforcement learning control module. The reinforcement learning control module trains the control strategy based on a pre-built electro-thermal coupled multiphysics model. The electro-thermal coupled multiphysics model is used to simulate the temperature field evolution, electrode distribution, pulse energy input and thermal coupling process of the reaction cavity. Step S3: The reinforcement learning control module infers or retrains online based on the environmental state and reward function, and dynamically generates a multivariate control strategy for pulse voltage amplitude, pulse width and pulse frequency; Step S4: Convert the control strategy into a control signal and apply it to the power supply and the multi-point electrode array to achieve coordinated regulation of the input energy of the multi-point electrodes; Step S5: Through the control process of steps S1-S4, the temperature field of the reaction chamber is precisely controlled and uniformly heated, thereby improving the stability and adaptability of the Joule flash evaporation process under complex thermal fields.

[0013] Preferably, the reinforcement learning control module obtains positive performance evaluation when the temperature deviation decreases, the temperature distribution uniformity improves, and the energy utilization rate improves, so as to realize the system's adaptive optimization and steady-state temperature control performance improvement. The positive performance evaluation includes the following reward function: , in, Indicates average temperature. Indicates the target temperature. Indicates the temperature gradient index, Indicates energy consumption-related indicators, This indicates the change in the control variable within adjacent control periods. This indicates the degree of dispersion in the temperature distribution at multiple points. , , , , These are the weighting coefficients.

[0014] Preferably, the modular reaction chamber structure and multi-point electrode array support large-scale expansion to improve material preparation volume and system scalability.

[0015] Preferably, the reinforcement learning control module applies the electrical energy output from the pulse power supply to the multi-point electrode array, thereby achieving coordinated regulation of the input energy of each electrode, which in turn achieves uniform current distribution, reduces the temperature gradient in the reaction chamber, and improves the consistency of product structure and the stability of the preparation process.

[0016] Through the above scheme, the present invention can realize multivariable intelligent temperature control and dynamic energy distribution optimization in the Joule thermal flash evaporation process, thereby overcoming the limitations of traditional single-variable constant drive control.

[0017] This invention is applicable to the efficient preparation of nanomaterials, functional carbon materials, and biomass-derived materials, and the provided method has at least the following beneficial effects: (1) Significantly improved large-scale preparation capability: The modular reaction chamber design and scalable electrode array structure support continuous or batch preparation of gram-level to kilogram-level samples to meet industrialization needs; (2) Enhanced temperature field uniformity: Through multi-point electrode synergistic heating, dynamic current distribution adjustment and intelligent temperature control strategy, the local current density fluctuation can be effectively reduced, making the thermal field more uniform and stable, thereby significantly reducing the temperature gradient inside the sample and improving the consistency of material processing.

[0018] (3) Multi-parameter coordinated control to achieve high-precision energy input: By using the linkage control of variables such as pulse amplitude, width and frequency, the energy input mode can be flexibly configured; this design can avoid adverse effects such as local overheating and thermal shock, and further improve the temperature control accuracy and energy utilization efficiency.

[0019] (4) Intelligent adaptive control: The reinforcement learning algorithm is introduced to automatically update the control strategy based on real-time sensor feedback, so that the system can maintain stable operation under different working conditions; the self-learning ability significantly improves the stability, repeatability and control robustness of the heating process.

[0020] (5) The device has a simple structure and strong expandability: The system adopts a modular structure design, which is convenient for disassembly and maintenance. At the same time, the number of electrodes, reaction chamber size and control modules can be flexibly expanded according to different process requirements. It is suitable for industrial applications of carbon material graphitization, biochar preparation, metal oxide reduction and other high-temperature heat treatment processes. Attached Figure Description

[0021] To more clearly illustrate the technical solution of the present invention, the present invention will be further described below with reference to the accompanying drawings. It should be understood that the accompanying drawings are only used to illustrate embodiments of the present invention and do not constitute a limitation on the scope of protection of the present invention.

[0022] Figure 1 This is a schematic diagram of the structure of a reinforcement learning-driven intelligent temperature control Joule heating amplification device provided in an embodiment of the present invention.

[0023] Figure 2 A functional structure diagram of the reinforcement learning control module provided in an embodiment of the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0025] Example 1 like Figure 1 As shown in the figure, this embodiment provides a reinforcement learning-driven intelligent temperature control Joule heating amplification device. Through the modular reaction chamber structure, multi-point electrode array arrangement and reinforcement learning intelligent control algorithm design, it realizes multi-dimensional collaborative optimization control from power input to temperature distribution, which improves system scalability and energy efficiency while ensuring heating uniformity.

[0026] This embodiment provides a reinforcement learning-driven intelligent temperature-controlled Joule heating amplification device. Through a modular reaction chamber structure, multi-point electrode array arrangement, and intelligent control algorithm design, this device achieves multi-dimensional collaborative optimization control from electrical energy input to temperature distribution, improving system scalability and energy efficiency while ensuring heating uniformity. It includes a power supply and control unit, an energy storage and pulse modulation unit, a reaction chamber unit, a multi-point electrode array unit, and a temperature monitoring and intelligent control unit connected in sequence.

[0027] The power supply and control unit includes a constant voltage power supply module and a pulse signal control module. It can output a constant voltage or a pulse voltage signal with adjustable amplitude, pulse width, and pulse frequency to drive a multi-point electrode array for rapid heating, providing an adjustable pulse voltage output signal. The amplitude, pulse width, and pulse frequency of this pulse voltage signal can all be adjusted by the reinforcement learning control module. The power supply module adopts an isolated design, supports high-frequency pulse output, and has overcurrent and overvoltage protection functions to meet the requirements of rapid energy injection and safety in the Joule thermal flash evaporation process.

[0028] The energy storage and pulse modulation unit consists of a capacitor energy storage module, a MOSFET or IGBT power switch module, and a filtering unit. It is used to precisely modulate the amplitude, pulse width, and pulse frequency of the pulse signal to achieve flexible control of the energy input mode. The power switch module uses PWM or a full-bridge topology to modulate the pulse amplitude, enabling different energy input strategies.

[0029] The reaction chamber unit adopts a modular chamber structure, and the chamber size can be enlarged or reduced according to the material preparation volume requirements. It can also be expanded into single-chamber or multi-chamber parallel connections through quick-release interfaces to meet the needs of different preparation scales. The chamber is equipped with multiple temperature sensor interfaces and multiple electrode mounting holes to ensure that functionality can be expanded without changing the main structure.

[0030] A multi-point electrode array unit consists of multiple graphite or metal electrodes evenly distributed along the sample area. Each electrode is connected to an independent, controllable channel, forming a scalable multi-point heating structure that effectively reduces the temperature gradient and improves the uniformity of the temperature field. Through multi-point Joule heating, distributed electrical energy input in space is achieved, reducing the risk of localized overheating of the sample and enhancing the overall uniformity of the temperature field.

[0031] The temperature monitoring and intelligent control unit comprises a multi-point thermocouple sensor array, a data acquisition module (DAQ), a PLC controller, and a reinforcement learning algorithm module (RL Agent). The reinforcement learning control module uses multi-point temperature signals as environmental feedback and, through interaction with system state, actions, and reward functions, adaptively optimizes the amplitude, width, and frequency of pulse voltages to achieve dynamic and coordinated adjustment of the multi-point electrode input voltages, thereby obtaining stable, uniform, and efficient temperature field control. This unit acquires, processes, and intelligently controls the real-time temperature data within the reaction chamber.

[0032] like Figure 2 As shown, the reinforcement learning control module proposed in this embodiment uses multi-point temperature signals as environmental feedback. Through the interaction between the signal and the system state, control actions, and reward function, it adaptively optimizes the amplitude, width, and frequency of the pulse voltage to achieve dynamic and coordinated adjustment of the input energy of the multi-point electrodes. This is applicable to the modular reaction chamber and scalable electrode array structure described in this embodiment. The system achieves highly uniform, efficient, and stable temperature control for large-size samples through multi-point temperature acquisition, intelligent strategy optimization, and coordinated pulse energy regulation.

[0033] Furthermore, this embodiment also provides a temperature regulation method based on a combination of the COMSOL-Simulink multiphysics co-simulation platform and a reinforcement learning (RL) intelligent control strategy. This method utilizes a simulation-control co-modeling framework to achieve real-time prediction, strategy optimization, and closed-loop intelligent regulation of the temperature field evolution process.

[0034] This embodiment also provides a method for a reinforcement learning-driven intelligent temperature control Joule heating amplification device, the steps of which are as follows: Step S1: Temperature Data Acquisition and Status Construction Multiple temperature sensors (multi-point thermocouples (T1~T1)) are arranged inside the reaction chamber. n Real-time monitoring of sample temperature is performed. The PLC controller reads the temperature signal at a set sampling period (e.g., 10 ms to 100 ms), and after filtering and normalization, forms the system state vector. in, The average temperature. The standard deviation of space temperature. This represents the peak temperature difference, used to reflect the uniformity of heating.

[0035] in A three-dimensional transient multiphysics coupled model of the sample, graphite electrode, and reaction chamber was constructed in COMSOL. The Joule heating module, the electro-thermal coupling solution module, and the transient temperature field solver were integrated into the same simulation project to form a digital twin model that can describe the thermal process of the actual device. The key node parameters of the temperature field were mapped and output in real time through the interface between COMSOL and Simulink (Livelink mechanism). A temperature data buffer and transmission module was built in Simulink, and a data interaction channel was established between MATLAB Engine and Python environment to realize the real-time acquisition and synchronous output of multi-point temperature data.

[0036] Step S2: Reinforcement Learning Control Decision The state vector formed in step S1 is input into the reinforcement learning control module (RL intelligent control module); the module trains the control strategy based on a pre-built electro-thermal coupled multiphysics model (COMSOL-Simulink co-simulation model) to simulate the temperature field evolution, electrode distribution, pulse energy input and thermal coupling process of the reaction cavity; Based on state input Output control actions through the policy network The control variables are pulse voltage amplitude, pulse width, and pulse frequency. The control strategy employs the Proximal Policy Optimization (PPO) algorithm to achieve continuous motion space optimization.

[0037] A reinforcement learning intelligent controller is built on the Python side based on the Stable-Baselines3 framework. The controller includes an environment interface module, a policy network module, a value network module, and a reward function module. Through the MATLAB Engine API, the current temperature, temperature gradient, heating power, and other state variables output by Simulink are transmitted to the Python side in real time as state inputs for the reinforcement learning agent. The reinforcement learning controller evaluates the policy based on the current state according to the multi-objective reward function and generates the control action for the next moment.

[0038] Step S3, Action Execution and Energy Modulation The control signal output by the RL intelligent control module is transmitted to the pulse modulation and power drive module via the PLC, and then acts on the expandable multi-point electrode array to achieve precise current distribution adjustment of the input energy of each electrode.

[0039] When the temperature is too low or the heating rate is insufficient, the RL intelligent control module can automatically increase the pulse amplitude or width. When local overheating or excessive temperature gradient is detected, the RL intelligent control module can reduce the amplitude and increase the frequency to achieve balanced energy distribution.

[0040] The policy network output of the reinforcement learning agent includes multiple continuous control variables such as pulse voltage amplitude, pulse width, pulse frequency, and duty cycle, which constitute the heating control action vector of the multiple input multiple output (MIMO) system. The Python side writes the control actions into the power module and electrode drive module of Simulink in real time through MATLAB Engine, so that the transient current density in COMSOL and the Joule heating process can be dynamically updated.

[0041] Step S4: Environmental Feedback and Reward Calculation After each control cycle, the system calculates a reward value based on real-time temperature and energy consumption information. This reward value is used to achieve adaptive optimization and improve steady-state temperature control performance. The reward function comprehensively considers temperature accuracy, heating uniformity, and energy efficiency, and is expressed as: in, For the target temperature, For energy consumption items, It characterizes the change amplitude of pulse voltage regulation at adjacent control moments, and is used to constrain the stability of the control process and the safety of system operation. This characterizes the statistical dispersion of temperature at multiple points within the reaction chamber, and is used to further measure the consistency of the temperature field distribution. These are weighting coefficients used to balance the relative influence of different performance indicators in the reward function. The reward function comprehensively considers temperature accuracy, heating uniformity, and energy efficiency.

[0042] Step S5: Policy Update and Adaptive Optimization The reinforcement learning module updates the control strategy parameters based on reward signals, achieving self-learning and self-optimization of the heating mode through repeated iterations. The control process can be rapidly transferred and retrained online under different material and process conditions, thereby maintaining precise temperature control, heating uniformity, product structure consistency, and process stability in the Joule flash evaporation process.

[0043] The control system can achieve rapid migration and online retraining under different material and process conditions, thereby maintaining optimal control performance over the long term.

[0044] A pulsed power supply and a multi-point electrode array apply voltage in real time according to the parameters output by the reinforcement learning intelligent controller, generating transient Joule heat with controllable spatial distribution inside the sample; the COMSOL solver calculates the updated temperature distribution based on the new electrical input conditions; Simulink transmits the temperature field data back to the Python environment, forming a closed-loop feedback; the reinforcement learning agent continuously iterates and optimizes the control strategy based on the closed-loop feedback, improving the accuracy of temperature regulation, heating efficiency, and temperature field uniformity.

[0045] Example 2 Comparison of the structural properties of biochar prepared by traditional temperature control and reinforcement learning temperature control This embodiment compares the structural performance differences of biochar prepared using traditional PID temperature control and reinforcement learning intelligent temperature control, under the same device structure, raw material conditions, and process parameters.

[0046] Step 1: Comparative Experiment The experiment used the reinforcement learning-driven intelligent temperature-controlled Joule heating amplification device of Example 1. The reaction chamber volume, number of electrodes and arrangement were kept consistent. The biomass raw materials used were agricultural waste from the same batch after pretreatment, with the same particle size distribution and initial moisture content.

[0047] The following two temperature control methods were used during the preparation process: Comparative Example: PID Temperature Control The traditional temperature closed-loop control method based on proportional-integral-derivative (PID) is adopted. The temperature signal collected by a single point or a small number of temperature sensors in the reaction chamber is used as the feedback input. The amplitude or frequency of the pulse voltage is adjusted by the PID parameters tuned by human experience in order to achieve the tracking control of the target temperature.

[0048] In this control method, the PID parameters remain unchanged during device operation. The thermal coupling relationship between multiple electrodes is not modeled or optimized, making it difficult to coordinate and regulate the overall temperature field in the reaction chamber under the condition that multiple heating points are working simultaneously.

[0049] Example: The reinforcement learning temperature control adopts the reinforcement learning-driven intelligent temperature control method for multi-point electrodes described in Example 2. Based on the real-time temperature information collected by multi-point temperature sensors arranged near each heating point, the multi-dimensional temperature state is input into the trained reinforcement learning controller, which adaptively adjusts the pulse voltage parameters corresponding to each heating point to achieve dynamic coordinated heating and temperature field equalization control among multi-point electrodes.

[0050] Except for the different temperature control methods, all other process parameters are kept the same. The target temperature range for the Joule flash evaporation process is 900℃, and the heating and holding times are the same.

[0051] Step 2: Biochar Preparation and Testing Methods After completing the Joule thermal flash evaporation reaction and cooling, biochar samples prepared under two different control methods were collected. The samples were dried, and their pore structure properties were characterized using a nitrogen adsorption-desorption method. The test parameters included specific surface area, total pore volume, and average pore size.

[0052] The structural properties of biochar prepared under different temperature control methods are shown in Table 1.

[0053] Table 1 Comparison of structural properties of biochar under different temperature control methods As shown in Table 1, under the same device structure and process conditions, biochar prepared using reinforcement learning temperature intelligent control has a significantly higher specific surface area than biochar prepared using traditional PID temperature control. Simultaneously, the total pore volume is significantly increased, indicating the formation of a larger and more densely distributed pore structure within the material. Further analysis reveals that biochar prepared under traditional PID temperature control has a larger average pore size but a lower total pore volume, indicating that the pore structure formed under this condition is dominated by a small number of large or irregular pores, with a limited overall pore structure development. In contrast, biochar prepared using reinforcement learning temperature intelligent control has a relatively smaller average pore size, but a significantly increased specific surface area and total pore volume, indicating the formation of a multi-level pore structure system dominated by micropores and mesopores, with a large number of uniformly distributed pores, which is more conducive to increasing the specific surface area.

[0054] Analysis suggests that the difference primarily stems from the varying capabilities of different temperature control methods in regulating the uniformity of the temperature field within the reaction chamber. Traditional PID temperature control mainly adjusts local temperatures, which is insufficient to effectively suppress temperature gradients under multi-point electrode Joule heating conditions, easily leading to localized overheating and causing partial collapse or coarsening of the pore structure. In contrast, reinforcement learning-based intelligent temperature control can collaboratively optimize the control of each heating point based on multi-point temperature feedback, significantly reducing temperature field fluctuations. This allows biomass to undergo a more uniform and controllable thermal process during pyrolysis, thereby promoting the stable formation of micropores and mesopores and significantly improving the overall pore structure performance of biochar.

[0055] This embodiment demonstrates that, in the application scenario of preparing biochar by multi-point electrode Joule thermal flash evaporation, compared with the traditional PID temperature control method, the reinforcement learning-driven intelligent temperature control method of this invention can achieve more effective temperature field coordinated regulation under complex and strongly coupled thermal field conditions, thereby bringing about a significant improvement in the structural performance of biochar.

[0056] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A reinforcement learning-driven intelligent temperature-controlled Joule heating amplification device, characterized in that, It includes a power supply and control unit, an energy storage and pulse modulation unit, a reaction chamber unit, a multi-point electrode array unit, and a temperature monitoring and intelligent control unit; The power supply and control unit includes a constant voltage power supply module and a pulse power supply module, which are used to output a constant voltage or a pulse voltage signal with adjustable amplitude, width and frequency. The energy storage and pulse modulation unit includes a capacitor energy storage module, a power switch module, and a filter unit, used to modulate the amplitude, width, and frequency of the pulse voltage; The reaction chamber unit adopts a modular and expandable structure, and the chamber size can be enlarged or reduced according to the preparation requirements. The multi-point electrode array unit is distributed with multiple graphite or metal electrodes along the inside of the reaction chamber to achieve multi-point heating of the sample and reduce the temperature gradient. The temperature monitoring and intelligent control unit includes a multi-point temperature sensor, a PLC controller, and a reinforcement learning control module; the multi-point temperature sensor is used to collect temperature data of multiple heating points in the reaction chamber in real time. The reinforcement learning control module adaptively adjusts the input voltage of the multi-point electrodes based on the feedback signals from the multi-point temperature sensors, thereby achieving dynamic and coordinated control of the temperature field of the reaction chamber.

2. The reinforcement learning-driven intelligent temperature-controlled Joule heating amplification device according to claim 1, characterized in that, The multi-point temperature sensors employ distributed thermocouples or infrared temperature sensors to collect temperature data from each heating point within the reaction chamber in real time, forming multi-dimensional temperature status data. This data is then transmitted to the central control system, where a reinforcement learning control module generates optimized control signals to achieve closed-loop temperature regulation.

3. A reinforcement learning-driven intelligent temperature-controlled Joule heating amplification device according to claim 1 or 2, characterized in that, The reinforcement learning algorithm module adaptively optimizes the amplitude, width, and frequency of the pulse voltage based on the interaction between the environmental state, control actions, and reward function, so as to achieve high-precision control of the temperature field.

4. A reinforcement learning-driven intelligent temperature-controlled Joule heating amplification device according to any one of claims 1 to 3, characterized in that, The reaction chamber unit is connected in series or in parallel with multiple reaction chamber units through a modular interface to expand the production capacity and meet the production needs of different scales.

5. A reinforcement learning-driven intelligent temperature-controlled Joule heating amplification device according to any one of claims 1 to 4, characterized in that, The number of electrodes, electrode spacing, electrode arrangement, and electrode material of the multi-point electrode array are adjusted according to the sample properties and heat treatment process requirements to optimize the uniformity of current and temperature distribution.

6. A method for a reinforcement learning-driven intelligent temperature-controlled Joule heating amplification device based on any one of claims 1 to 5, characterized in that, Includes the following steps: Step S1: Collect temperature data inside the reaction chamber in real time using multi-point temperature sensors; Step S2: Input the temperature data into the reinforcement learning control module. The reinforcement learning control module trains the control strategy based on a pre-built electro-thermal coupled multiphysics model. The electro-thermal coupled multiphysics model is used to simulate the temperature field evolution, electrode distribution, pulse energy input and thermal coupling process of the reaction cavity. Step S3: The reinforcement learning control module infers or retrains online based on the environmental state and reward function, and dynamically generates a multivariate control strategy for pulse voltage amplitude, pulse width and pulse frequency; Step S4: Convert the control strategy into a control signal and apply it to the power supply and the multi-point electrode array to achieve coordinated regulation of the input energy of the multi-point electrodes; Step S5: Through the control process of steps S1-S4, the temperature field of the reaction chamber is precisely controlled and uniformly heated, thereby improving the stability and adaptability of the Joule flash evaporation process under complex thermal fields.

7. The method for a reinforcement learning-driven intelligent temperature control Joule heating amplification device according to claim 6, characterized in that, The reinforcement learning control module obtains positive performance evaluation when the temperature deviation decreases, the temperature distribution uniformity improves, and the energy utilization rate improves, which is used to realize system adaptive optimization and steady-state temperature control performance improvement. The positive performance evaluation includes the following reward function: ， in, Indicates average temperature. Indicates the target temperature. Indicates the temperature gradient index, Indicates energy consumption-related indicators, This indicates the change in the control variable within adjacent control periods. This indicates the degree of dispersion in the temperature distribution at multiple points. , , , , These are the weighting coefficients.

8. A method for a reinforcement learning-driven intelligent temperature-controlled Joule heating amplification device according to any one of claims 6 or 7, characterized in that, The modular reaction chamber structure and multi-point electrode array support large-scale expansion, which can improve the material preparation volume and system scalability.

9. The method for a reinforcement learning-driven intelligent temperature control Joule heating amplification device according to claim 8, characterized in that, The reinforcement learning control module applies the electrical energy output from the pulse power supply to the multi-point electrode array, thereby achieving coordinated regulation of the input energy of each electrode. This results in uniform current distribution, reduced temperature gradient within the reaction chamber, and improved product structure consistency and stability of the preparation process.

Images