Intelligent regulation and control method and system suitable for SF6 waste gas microwave plasma treatment
By constructing a particle and energy balance model, combining a neural ODE model and a multi-agent reinforcement learning algorithm, and adjusting microwave power and gas flow rate in real time, the problems of non-uniform power distribution and prediction error in microwave plasma reaction systems were solved, and the generation of byproducts was suppressed and the decomposition efficiency was improved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-19
- Publication Date
- 2026-04-17
AI Technical Summary
Existing microwave plasma reaction systems fail to effectively reflect the complex relationship between gas flow rate, temperature gradient, and reaction rate over time, leading to amplified prediction errors. They are unable to provide inhibitory intervention in the early stages of byproduct formation, and the non-uniform power distribution results in the enrichment of byproducts in local overheated reaction zones, making it difficult to achieve coordinated optimization.
By constructing a particle and energy balance model, the fading effect of the microwave propagation path is analyzed. Combining a neural ODE model and a multi-agent reinforcement learning algorithm, the microwave power and gas flow rate are adjusted in real time. A centralized critic algorithm is used for error correction and parameter updates to achieve active regulation of the reaction zone.
Precise control of the microwave plasma reaction system has been achieved, reducing the generation of byproducts, improving decomposition efficiency and purity, and ensuring stable operation of the system under different operating conditions.
Smart Images
Figure CN121869023A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent control technology, and in particular to an intelligent control method and system suitable for microwave plasma treatment of SF6 waste gas. Background Technology
[0002] SF6 has extremely high chemical stability and extremely low ionization energy, making it extremely difficult to decompose in the natural environment. It has become a key target in greenhouse gas control. To improve treatment efficiency, microwave plasma technology has attracted attention because it can achieve high-energy electron-molecule collision decomposition at normal pressure. This technology uses a microwave field to excite plasma in the reaction chamber and decomposes SF6 into low-fluoride compounds and fluorine free radicals through high temperature and high-energy electron action. Existing microwave plasma reaction systems often suffer from complex reaction product structures and byproducts such as OF2, SO2, and F2, which are toxic and corrosive. This not only reduces decomposition efficiency but may also damage the cavity walls and electrodes. Current SF6 treatment ignores the multi-scale fading effect of the microwave propagation path, resulting in strong spatial non-uniformity in power distribution. This leads to local overheated reaction zones becoming byproduct accumulation points. Furthermore, it fails to reflect the complex relationship between gas flow rate, temperature gradient, and reaction rate over time, amplifying prediction errors and making it impossible to suppress byproduct formation in its early stages, thus hindering coordinated optimization. Summary of the Invention
[0003] In view of the above-mentioned existing problems, the present invention provides an intelligent control method and system for microwave plasma treatment of SF6 waste gas. This solves the problems of existing SF6 treatments that ignore the multi-scale fading effect of microwave propagation path, and the strong non-uniformity of power distribution in space, which leads to local overheated reaction zones becoming by-product accumulation points. Furthermore, it fails to reflect the complex relationship between gas flow rate, temperature gradient and reaction rate over time, resulting in amplified prediction errors, making it impossible to suppress intervention in the early stage of by-product formation, and making it difficult to achieve coordinated optimization.
[0004] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides an intelligent control method for microwave plasma treatment of SF6 waste gas, which includes dividing the SF6 waste gas treatment cavity into regions, constructing a design particle and energy balance model, updating the particle concentration and energy density, analyzing free space path loss to calculate the large-scale fading coefficient, analyzing random fluctuations in microwave propagation to calculate the small-scale fading coefficient, and then calculating the comprehensive attenuation, and constructing a connected beamforming receiving signal model in combination with the distribution of microwave power to determine the received signal strength. Based on the received signal strength, the external source power is calculated, a temperature change model is established to calculate the temperature change, a neural ODE model is constructed, the convection term and net reaction rate of gas inflow and outflow and SF6 concentration are analyzed, and the external source power, SF6 concentration and temperature data are combined into a time feature vector. Linear regression is used as the residual term, and the right-hand function is constructed by combining the convection term and net reaction rate. The RK4 algorithm is used to calculate the predicted value of SF6 concentration. Based on the Multi-Agent Reinforcement Learning (MARL) model, the intelligent agent is responsible for regulating the agent parameters and action space. The Deep Deterministic Policy Gradient (DDPG) algorithm is used to update the agent parameters and calculate the reward. Combined with the neural ODE model, the centralized critic algorithm is used to calculate the temporal difference error and iteratively update the action space parameters. Save the data and generate log data for cloud storage.
[0005] As a further preferred embodiment, the calculation of comprehensive attenuation, combined with the distribution of microwave power to construct a connected beamforming received signal model to determine the received signal strength, includes: Design particle and energy balance models are constructed for each region, including the particle balance equation for the reaction chamber, which describes the change of SF6 concentration over time; the energy balance equation for the gas inlet, which describes the change of energy over time within the reaction chamber; and the mass balance equation for the gas outlet. The particle balance equation for the reaction chamber and the energy balance equation for the gas inlet are discretized in time using the Euler method, and the particle concentration and energy density are updated respectively. The Euclidean distance is calculated based on the spatial relationship between the receiving point and the microwave source. The large-scale fading coefficient is calculated based on the free space path loss model. The small-scale fading coefficient is calculated by analyzing the random fluctuations in the microwave propagation process and the wavelength of the microwave signal. The comprehensive attenuation coefficient of the signal propagation path and physical environment conditions is analyzed by combining the large-scale fading coefficient. Based on the SF6 concentration and temperature data of each region, and considering that the attenuation and propagation of microwave signals directly affect the changes in SF6 concentration in the reaction chamber during microwave processing, a connected beamforming receiving signal model is constructed using microwave power distribution and comprehensive attenuation coefficient to determine the received signal strength.
[0006] As a further preferred embodiment, the construction of the neural ODE model analyzes the convection term and net reaction rate of gas inflow and outflow rates and SF6 concentration. It then uses external source power, SF6 concentration, and temperature data to form a time feature vector. Linear regression is used as the residual term, and the right-hand side function is constructed by combining the convection term and net reaction rate. The predicted SF6 concentration is calculated using the RK4 algorithm, including: Based on the received signal strength, the proportionality constant is determined using historical data, and the power of the external source is calculated. A temperature change model is established based on the energy conservation equation. Temperature data is calculated based on the density of the reacting gas, the specific heat capacity of the gas at constant pressure, and the heat loss power density. The temperature change is then discretized. A neural ODE model was constructed based on the particle balance equation and the energy balance equation. The convection term and net reaction rate were analyzed by measuring the gas inlet and outlet flow rates and SF6 concentration in the reaction chamber area. The time feature vector is composed of data including external source power, SF6 concentration, and temperature. The target derivative is set at each time step, and the parameter vector is calculated by the least squares method. The feature vector is used as the residual term through linear regression, and the right-hand function is constructed by combining the convection term and the net reaction rate. The predicted SF6 concentration is calculated by performing a one-step integration using the RK4 algorithm based on the right-hand side function.
[0007] As a further preferred embodiment, the Multi-Agent Reinforcement Learning (MARL) defines the agent parameters and action space that the intelligent agent is responsible for regulating, and uses the Deep Deterministic Policy Gradient (DDPG) algorithm to update the agent parameters, including: Based on the Multi-Agent Reinforcement Learning (MARL) algorithm, the agent parameters that the intelligent agent is responsible for regulating are defined, including controlling microwave power, controlling gas flow rate, and controlling SF6 concentration stability. The state space of the agent parameters is defined, including microwave power, temperature, concentration, and gas flow rate. The action space is defined, including the adjustment amount of microwave power and the change amount of gas flow rate. The safety boundary is determined based on historical experience. The Deep Deterministic Policy Gradient (DDPG) algorithm was selected for training, and state updates were performed based on the agent parameters, with rewards calculated based on the concentration error.
[0008] As a further preferred embodiment, the adoption of a centralized commentator algorithm to calculate the temporal difference error and iteratively update the action space parameters includes: The agent's policy network is input with agent parameters and action space, and the neural ODE model is input to determine SF6 concentration and temperature data. Rewards are calculated in a timely manner. A centralized critic algorithm is used to calculate the temporal difference error by forming quadruples based on agent parameters and action space parameters. Gradient descent calculation is performed with the goal of minimizing the mean square error. The surrogate parameters and action space parameters are updated. When the average concentration error is less than or equal to the convergence threshold, the update of the output action space parameters is stopped as the optimal control strategy.
[0009] As a further preferred embodiment, the step of saving data and generating log data for cloud storage includes: The output action space parameters and agent parameters are saved, logs are generated using the ELK Stack tool, encrypted with SSL / TLS communication, and stored in the cloud.
[0010] As a further preferred embodiment, the division of the SF6 exhaust gas treatment chamber into zones includes: The microwave treatment cavity for SF6 waste gas treatment is divided into zones, including a reaction chamber, a gas inlet, and a gas outlet.
[0011] Secondly, the present invention provides an intelligent control system suitable for microwave plasma treatment of SF6 waste gas, comprising: The equilibrium modeling module is used to establish and discretely update particle and energy balance equations according to cavity regions. The fading modeling module is used to calculate the spatial distribution of field strength based on the combined attenuation characterization of geometric distance and free space path loss superimposed with stochastic small-scale effects. The beamforming receiving signal module is used to couple multi-source power distribution with integrated attenuation and output the received signal strength at each measurement point; The external source power estimation module is used to calibrate the proportionality constant using historical data to achieve online mapping from received strength to external source power; The energy model module is used to model and discretely solve the temperature based on energy conservation and heat loss, providing the time-series changes of temperature at multiple measurement points. The Neural ODE Dynamics module is used to integrate temporal features such as concentration, temperature, and external power to construct state differential equations with physical priors. The rate assessment module is used to combine inflow and outflow rates with concentration field to calculate convection term and net reaction rate, characterizing mass transfer and reaction coupling effect. The multi-agent module is used to define the state, action space, and safety boundary, and to make decisions on power and flow rate respectively to coordinate and stabilize the concentration. The reward assessment module is used to comprehensively consider concentration error, energy consumption, and temperature / concentration safety penalties. The data and log subsystem is used to persist critical states and control variables, generate searchable logs, and securely upload them to the cloud.
[0012] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the intelligent control method for microwave plasma treatment of SF6 waste gas as described in the first aspect of the present invention.
[0013] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the intelligent control method for microwave plasma treatment of SF6 waste gas as described in the first aspect of the present invention.
[0014] The beneficial effects of this invention are as follows: By calculating the comprehensive attenuation coefficient, random fluctuations and physical propagation conditions are considered simultaneously, enabling the system to accurately reflect microwave transmission characteristics under different operating conditions. By learning the true evolution trend of the reactant gas through continuous-time differential equations, the nonlinear fluctuations in temperature and concentration caused by changes in microwave input, gas flow, and discharge intensity are captured. Through a centralized commentator algorithm, the intelligent agent can simultaneously consider multiple factors such as SF6 concentration, temperature, flow rate, and energy consumption. When a side reaction trend indicated by the prediction model is detected, the microwave power distribution and gas flow rate are automatically adjusted to achieve active regulation of energy and material transport in the reaction zone. Attached Figure Description
[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0016] Figure 1 This is a flowchart illustrating the intelligent control method for microwave plasma treatment of SF6 waste gas in Example 1. Figure 2 This is a schematic diagram of the intelligent control system for microwave plasma treatment of SF6 waste gas in Example 1; Figure 3 This is a schematic diagram of the RK4 integral of the intelligent control method applicable to microwave plasma treatment of SF6 waste gas in Example 1; Figure 4 This is a schematic diagram of multi-agent reinforcement learning for the intelligent control method applicable to microwave plasma treatment of SF6 waste gas in Example 1. Detailed Implementation
[0017] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
[0018] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.
[0019] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.
[0020] Example 1, referring to Figures 1 to 4 This is the first embodiment of the present invention, which provides an intelligent control method suitable for microwave plasma treatment of SF6 waste gas, including the following steps: S1. For the SF6 exhaust gas treatment cavity, the area is divided, a particle and energy balance model is constructed, the particle concentration and energy density are updated, the free space path loss is analyzed to calculate the large-scale fading coefficient, and the random fluctuations in the microwave propagation process are analyzed to calculate the small-scale fading coefficient. Then, the comprehensive attenuation is calculated, and the connected beamforming receiving signal model is constructed in combination with the microwave power distribution to determine the received signal strength. Preferably, the microwave treatment cavity for SF6 waste gas treatment is divided into zones, including a reaction chamber, a gas inlet, and a gas outlet.
[0021] Furthermore, the overall attenuation is calculated, and a connected beamforming received signal model is constructed based on the microwave power distribution to determine the received signal strength, including: Based on the design of each region, particle and energy balance models are constructed, including the particle balance equation for the reaction chamber, which describes the change of SF6 concentration over time; the energy balance equation for the gas inlet, which describes the change of energy over time within the reaction chamber; and the mass balance equation for the gas outlet, expressed as follows:
[0022]
[0023]
[0024] in, This indicates the SF6 concentration in the reaction chamber area. This indicates the input of microwave energy from an external source. This indicates that the SF6 product generated during the reaction, This indicates SF6 transmitted from other areas. This indicates the energy density at the gas inlet. Indicates external auxiliary heating. Indicates the radiation loss of the gas. This indicates the energy transferred to other regions. This indicates the SF6 concentration at the gas outlet. This represents the SF6 flow rate at the export point. This indicates SF6 flowing out of the export area; The particle balance equations for the reaction chamber and the energy balance equations for the gas inlet are discretized over time using the Euler method, and updated for particle concentration and energy density respectively, as follows:
[0025]
[0026] in, This indicates the SF6 concentration at the next time step. Indicates the time step. , , as well as The values represent the SF6 concentration in the reaction chamber region at time t, the external microwave energy input, the SF6 products generated due to the reaction, and the SF6 transported from other regions, respectively. This indicates the energy density at the gas inlet at the next time step. , , as well as These represent the energy density at the gas inlet, external auxiliary heating, gas radiation loss, and energy transferred to other regions at time t, respectively. The Euclidean distance is calculated based on the spatial relationship between the receiving point and the microwave source. The large-scale fading coefficient is calculated using the free-space path loss model. The small-scale fading coefficient is calculated by analyzing random fluctuations during microwave propagation and the wavelength of the microwave signal. Finally, the comprehensive attenuation coefficient, considering the signal propagation path and physical environmental conditions, is analyzed using the large-scale fading coefficient and expressed as follows:
[0027]
[0028]
[0029]
[0030]
[0031] in, This represents the distance between the receiving point m and the microwave source k. , as well as This represents the coordinates of the receiving point m. , as well as This represents the coordinates of the receiving point k. Represents the large-scale fading coefficient. Represents the speed of light. This represents the small-scale fading coefficient. This represents random fluctuations during microwave propagation. The wavelength of a microwave signal. Indicates the frequency of the microwave signal. Indicates the overall attenuation coefficient; Based on the SF6 concentration and temperature data for each region, and considering that microwave signal attenuation and propagation directly affect the SF6 concentration changes within the reaction chamber during microwave processing, a beamforming receiver signal model is constructed using microwave power distribution and a comprehensive attenuation coefficient to determine the received signal strength, expressed as:
[0032]
[0033] in, The received signal strength is represented by , where represents the signal strength received at the m-th receiving point, and K represents the total number of microwave sources k. This represents the power of the k-th microwave source. The noise term at the receiving point m is typically determined by the noise of the equipment system. This represents the transmitted signal of the k-th microwave source. Indicates the amplitude of the microwave source signal. This indicates the frequency of the microwave source signal.
[0034] By constructing the particle balance equation of the reaction chamber, the energy balance equation of the gas inlet, and the material balance equation of the gas outlet, the system forms a closed-loop feedback at the level of microscopic particles and energy transfer. Through this coupling, the generation, consumption, and energy evolution of SF6 can be described simultaneously, and the dynamic consistency of concentration and temperature can be obtained. By synchronously updating particle concentration and energy density under Euler time discretization, a bivariate evolution with consistent time scale is achieved, enabling concentration changes to reflect differences in energy input in real time, thus solving the prediction lag problem caused by the asynchronous evolution of energy and matter in previous models. By introducing the Euclidean distance between the receiving point and the microwave source to establish a spatial coupling relationship, and combining the free space path loss and small-scale fading model, the local microwave field strength distribution in the reaction chamber is obtained. Through this spatial modeling, the non-uniformity of energy distribution and SF6 decomposition interval in the reaction chamber can be corrected in real time. By calculating the comprehensive attenuation coefficient, random fluctuations and physical propagation conditions are considered simultaneously, enabling the system to accurately reflect microwave transmission characteristics under different operating conditions and achieve adaptive compensation for complex electromagnetic environments.
[0035] S2, calculate the external source power, establish a temperature change model to calculate the temperature change, construct a neural ODE model, analyze the convection term and net reaction rate of gas inflow and outflow and SF6 concentration, and combine the external source power, SF6 concentration and temperature data into a time feature vector. Use linear regression as the residual term, combine the convection term and net reaction rate to construct the right-hand function, and use the RK4 algorithm to calculate the predicted value of SF6 concentration. Preferably, a neural ODE model is constructed to analyze the convection term and net reaction rate of gas inflow and outflow rates and SF6 concentration. The external source power, SF6 concentration, and temperature data are combined to form a time feature vector. Linear regression is used as the residual term, and the right-hand side function is constructed by combining the convection term and net reaction rate. The RK4 algorithm is then used to calculate the predicted SF6 concentration, including: Based on the received signal strength, the proportionality constant is determined using historical data, and the external source power is calculated, expressed as:
[0036]
[0037] in, This represents the proportionality constant, and N represents the total number of historical data. This represents the known transmit power of the i-th historical data point. This represents the signal strength measured at the m-th receiving point for the i-th historical data. The signal strength at time t is represented by the signal strength at time t. Indicates the power of the external source; A temperature change model is established based on the energy conservation equation. Temperature data is calculated using the reactant gas density, gas specific heat capacity at isobaric pressure, and heat loss power density. The temperature change is then discretized and expressed as follows:
[0038]
[0039] in, Indicates the density of the reacting gas. This represents the specific heat capacity of a gas at constant pressure. This represents the temperature data measured at the m-th receiving point. The heat loss power density is determined by heat dissipation measurements or cooling system design parameters. This represents the temperature data at the next time step, t. Indicates the time step. Indicates the effective volume of the reaction zone; A neural ODE model is constructed based on the particle equilibrium equation and the energy balance equation, and is expressed as follows:
[0040] The convection term and net reaction rate are analyzed based on the gas inflow and outflow rates and SF6 concentration in the reaction chamber region, and are expressed as follows:
[0041]
[0042] in, The convection term represents time t. and These represent the gas inflow and outflow rates during reaction time t, respectively. The concentration of SF6 in the reaction zone at time t. Indicates the effective volume of the reaction chamber. The term representing the net reaction rate at time t; The time feature vector is composed of data including external source power, SF6 concentration, and temperature. A target derivative is set at each time step, and the parameter vector is calculated using the least squares method. Based on the feature vector, linear regression is performed as the residual term. The right-hand side function is constructed by combining the convection term and the net reaction rate, and is expressed as:
[0043]
[0044]
[0045]
[0046]
[0047] in, Represents the time feature vector. Represents a parameter vector. Indicates transpose calculation. This represents a matrix composed of time eigenvectors at all times. This represents the objective derivative with respect to time. This indicates the calculation of the residual term; The predicted SF6 concentration is calculated by one-step integration using the RK4 algorithm based on the right-hand side function, and is expressed as follows:
[0048]
[0049]
[0050]
[0051]
[0052] in, Indicates the forward slope. Indicates the first-order slope. Indicates the second-order slope. Indicates the key slope. Indicates time The predicted value of SF6 concentration.
[0053] By establishing a proportional constant relationship between the received signal strength and historical transmit power data, the system can automatically calculate the power of the external source without external calibration, thereby achieving adaptive calibration and closed-loop calibration of microwave energy input. By coupling gas density, isobaric specific heat capacity, and heat loss power density through the energy conservation equation, temperature changes can truly reflect the dynamic balance between energy input and dissipation, thereby ensuring thermodynamic consistency in subsequent concentration predictions. By jointly constructing a neural ODE model using particle balance equations and energy balance equations, the reaction system can simultaneously learn chemical reaction kinetics and energy evolution characteristics, achieving a unified dynamic description of concentration and energy. By incorporating the gas inflow and outflow rates with the SF6 concentration into a convection term and introducing the net reaction rate, the model can not only describe the generation and consumption within the reaction zone, but also capture the transregional transport effects, thus avoiding local prediction bias. By constructing the right-hand side function by combining the linear regression residual term with the convection term and the net reaction rate, a dual correction based on physical constraints and data-driven approaches is achieved, making the model both interpretable and adaptive. By using the fourth-order Runge-Kutta integral algorithm on the right-hand function, the concentration prediction achieves high-order stability and accuracy in numerical terms, avoiding the oscillation error of the Euler integral in the highly nonlinear region.
[0054] S3, based on the Multi-Agent Reinforcement Learning (MARL) model, defines the agent parameters and action space that the intelligent agent is responsible for regulating. It uses the Deep Deterministic Policy Gradient (DDPG) algorithm to update the agent parameters and calculate rewards. Combined with the neural ODE model, it uses the centralized critic algorithm to calculate the temporal difference error and iteratively update the action space parameters. Preferably, based on Multi-Agent Reinforcement Learning (MARL), the intelligent agent is responsible for regulating the agent parameters and action space. The Deep Deterministic Policy Gradient (DDPG) algorithm is used to update the agent parameters, including: Based on the Multi-Agent Reinforcement Learning (MARL) algorithm, the agent parameters that the intelligent agent is responsible for regulating are defined, including controlling microwave power, controlling gas flow rate, and controlling SF6 concentration stability. The state space of the agent parameters is defined, including microwave power, temperature, concentration, and gas flow rate. The action space is defined, including the adjustment amount of microwave power and the change amount of gas flow rate. The safety boundary is determined based on historical experience. The Deep Deterministic Policy Gradient (DDPG) algorithm is selected for training. State updates are performed based on the agent parameters, and rewards are calculated according to the concentration error, as shown below:
[0055]
[0056]
[0057]
[0058]
[0059]
[0060] in, This represents the reward value at time t. This indicates the normalized SF6 concentration error. Indicates energy consumption weight. This represents the normalized value of microwave power. Indicates the weight of temperature exceeding the limit. This represents the temperature penalty value. Indicates concentration safety weight. This represents the security penalty value. This represents the historical mean of the error in the normalized SF6 concentration. This represents the normalized historical mean of microwave power.
[0061] By decomposing the multi-agent task into three controllable agents—"power, flow rate, and concentration stability"—and sharing a global reward that includes temperature and concentration safety factors, each agent is forced to be responsible for the overall system constraints while optimizing its own channel, thus reducing mutual interference and oscillations caused by local optimization. By adopting a deterministic strategy of DDPG to output continuous actions (power adjustment amount, flow rate change amount), it directly matches the continuously adjustable characteristics of the physical actuator, avoiding the quantization error caused by discretization and the thermal shock caused by frequent switching; By sharing key states and global rewards among multiple agents while each agent executes local actions, it is possible to track operating condition drift, feed fluctuations and changes in ambient temperature, thereby reducing the frequency of manual recalibration. Through a centralized commentator algorithm, the intelligent system can simultaneously consider multiple factors such as SF6 concentration, temperature, flow rate, and energy consumption. When it detects a side reaction trend indicated by the prediction model, it automatically adjusts the microwave power distribution and air intake flow rate to actively regulate the energy and material transport in the reaction zone, keeping the reaction in the main decomposition channel. This not only reduces the probability of byproduct formation but also pre-lowers the local energy density before side reactions occur, thereby maintaining high decomposition efficiency and high purity during dynamic operation.
[0062] Furthermore, a centralized commentator algorithm is employed to calculate the temporal difference error and iteratively update the action space parameters, including: The agent's policy network is input with agent parameters and action space, and SF6 concentration and temperature data are determined by inputting a neural ODE model. Rewards are calculated promptly. A centralized critic algorithm is used, and the temporal difference error is calculated based on a quadruple formed from agent parameters and action space parameters, expressed as:
[0063]
[0064] in, Indicates timing difference error. Indicates the discount factor. The expected return on investment can be determined based on historical experience using the ratio of SF6 flow rate to the effective volume of the reaction chamber, i.e., the engineering magnitude of the residence time. The Critic function represents the critic. and Let these represent the proxy parameters and action space parameter vectors for the next time step, respectively. and These represent the current agent parameter and action space parameter vectors, respectively. Gradient descent calculation is performed with the goal of minimizing the mean square error. The surrogate parameters and action space parameters are updated. When the average concentration error is less than or equal to the convergence threshold (determined based on historical experience), the update of the output action space parameters is stopped as the optimal control strategy. The output-based optimal control strategy includes action space parameters, namely the adjustment amount of microwave power and the change amount of gas flow rate. The parameters are adjusted by controlling the actuator as an indication. The microwave treatment cavity for SF6 waste gas treatment is adjusted. The control of the microwave power adjustment amount can deal with possible heat accumulation effects or by-product generation. The control of the change amount of gas flow rate can allow SF6 to have an appropriate residence time in the reaction zone, so as to fully decompose and improve the final decomposition rate. By coupling the input of the agent policy network with the SF6 concentration and temperature data output by the neural ODE model in real time, the state information of reinforcement learning not only contains controllable quantities, but also dynamically reflects the actual evolution of the response system, thereby making the policy training process physically consistent and real-time. By using agent parameters and action parameters as a quadruple input in a centralized commentator structure, the commentator can simultaneously evaluate the interaction between multiple agents when calculating the temporal difference error. This allows each agent to consider the overall collaborative performance of the system when updating its own policy, thus avoiding policy conflicts caused by traditional decentralized training. By introducing a discount factor and expected process return into the error calculation, the learning process can simultaneously take into account short-term response stability and long-term energy consumption optimization, forming a dynamic balance at the time level and improving the energy efficiency and stability of the strategy under continuous operation.
[0065] S4 saves data and generates log data for cloud storage. Preferably, the output action space parameters and agent parameters are saved as data, logs are generated using the ELK Stack tool, encrypted with SSL / TLS communication, and stored in the cloud.
[0066] This embodiment also provides an intelligent control system suitable for microwave plasma treatment of SF6 waste gas, including: Balance modeling module: Establishes and discretely updates particle and energy balance equations according to cavity regions; Fading modeling module: Based on the superposition of geometric distance and free space path loss with stochastic small-scale effects, calculate the comprehensive attenuation characterization of the spatial distribution of field strength; Beamforming receiver module: Couples multi-source power distribution with integrated attenuation to output the received signal strength at each measurement point; External source power estimation module: uses historical data to calibrate a proportional constant to achieve online mapping from received strength to external source power; Energy Model Module: Based on energy conservation and heat loss, model and solve discretely, providing the time-series changes of temperature at multiple measurement points; The neural ODE dynamics module integrates temporal features such as concentration, temperature, and external power to construct state differential equations with physical priors. Rate assessment module: Combines inflow and outflow rates with concentration field to calculate convection term and net reaction rate, characterizing mass transfer and reaction coupling effect; Multi-agent module: Defines state, action space and safety boundary, makes decisions on power and flow rate respectively to coordinate and stabilize concentration, and integrates concentration error, energy consumption and temperature / concentration safety penalties; Data and Log Subsystem: Persist critical states and control variables, generate searchable logs, and securely migrate them to the cloud.
[0067] This embodiment also provides a computer device applicable to the intelligent control method for microwave plasma treatment of SF6 waste gas, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the intelligent control method for microwave plasma treatment of SF6 waste gas as proposed in the above embodiment. The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.
[0068] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements the intelligent control method for microwave plasma treatment of SF6 waste gas as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.
[0069] In summary, this invention, by calculating the comprehensive attenuation coefficient, simultaneously considers random fluctuations and physical propagation conditions, enabling the system to accurately reflect microwave transmission characteristics under different operating conditions. It learns the true evolution trend of the reactant gas through continuous-time differential equations, capturing the nonlinear fluctuations in temperature and concentration caused by changes in microwave input, gas flow, and discharge intensity. Through a centralized commentator algorithm, the intelligent system simultaneously considers multiple factors such as SF6 concentration, temperature, flow rate, and energy consumption. When it detects a side reaction trend indicated by the prediction model, it automatically adjusts the microwave power distribution and inlet flow rate, achieving proactive regulation of energy and material transport in the reaction zone.
[0070] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A smart control method for microwave plasma treatment of SF6 waste gas, characterized in that, include: The SF6 exhaust gas treatment chamber is divided into regions, a particle and energy balance model is constructed, the particle concentration and energy density are updated, the free space path loss is analyzed to calculate the large-scale fading coefficient, and the random fluctuations in the microwave propagation process are analyzed to calculate the small-scale fading coefficient. Then, the comprehensive attenuation is calculated, and the beamforming receiving signal model is constructed in combination with the microwave power distribution to determine the received signal strength. Based on the received signal strength, the external source power is calculated, a temperature change model is established to calculate the temperature change, a neural ODE model is constructed, the convection term and net reaction rate of gas inflow and outflow and SF6 concentration are analyzed, and the external source power, SF6 concentration and temperature data are combined into a time feature vector. Linear regression is used as the residual term, and the right-hand function is constructed by combining the convection term and net reaction rate. The RK4 algorithm is used to calculate the predicted value of SF6 concentration. Based on the Multi-Agent Reinforcement Learning (MARL) model, the intelligent agent is responsible for regulating the agent parameters and action space. The Deep Deterministic Policy Gradient (DDPG) algorithm is used to update the agent parameters and calculate the reward. Combined with the neural ODE model, the centralized critic algorithm is used to calculate the temporal difference error and iteratively update the action space parameters. Save the data and generate log data for cloud storage.
2. The intelligent control method for microwave plasma treatment of SF6 waste gas as described in claim 1, characterized in that, The calculation of overall attenuation, combined with the microwave power distribution, constructs a connected beamforming received signal model to determine the received signal strength, including: Design particle and energy balance models are constructed for each region, including the particle balance equation for the reaction chamber, which describes the change of SF6 concentration over time; the energy balance equation for the gas inlet, which describes the change of energy over time within the reaction chamber; and the mass balance equation for the gas outlet. The particle balance equations for the reaction chamber and the energy balance equations for the gas inlet are discretized over time using the Euler method, and the particle concentration and energy density are updated accordingly. The Euclidean distance is calculated based on the spatial relationship between the receiving point and the microwave source. The large-scale fading coefficient is calculated based on the free space path loss model. The small-scale fading coefficient is calculated by analyzing the random fluctuations in the microwave propagation process and the wavelength of the microwave signal. The comprehensive attenuation coefficient of the signal propagation path and physical environment conditions is analyzed by combining the large-scale fading coefficient. Based on the SF6 concentration and temperature data of each region, and considering that the attenuation and propagation of microwave signals directly affect the changes in SF6 concentration in the reaction chamber during microwave processing, a beamforming receiving signal model is constructed using microwave power distribution and comprehensive attenuation coefficients to determine the received signal strength.
3. The intelligent control method for microwave plasma treatment of SF6 waste gas as described in claim 1, characterized in that, The neural ODE model is constructed to analyze the convection term and net reaction rate of gas inflow and outflow rates and SF6 concentration. External source power, SF6 concentration, and temperature data are combined to form a time feature vector. Linear regression is used as the residual term, and the right-hand side function is constructed by combining the convection term and net reaction rate. The RK4 algorithm is then used to calculate the predicted SF6 concentration, including: Based on the received signal strength, the proportionality constant is determined using historical data, and the power of the external source is calculated. A temperature change model is established based on the energy conservation equation. Temperature data is calculated based on the density of the reacting gas, the specific heat capacity of the gas at constant pressure, and the heat loss power density. The temperature change is then discretized. A neural ODE model was constructed based on the particle balance equation and the energy balance equation. The convection term and net reaction rate were analyzed by measuring the gas inlet and outlet flow rates and SF6 concentration in the reaction chamber area. The time feature vector is composed of data including external source power, SF6 concentration, and temperature. The target derivative is set at each time step, and the parameter vector is calculated by the least squares method. The feature vector is used as the residual term through linear regression, and the right-hand function is constructed by combining the convection term and the net reaction rate. The predicted SF6 concentration is calculated by performing a one-step integration using the RK4 algorithm based on the right-hand side function.
4. The intelligent regulation method for SF6 waste gas microwave plasma treatment of claim 1, wherein, The Multi-Agent Reinforcement Learning (MARL)-based approach defines the agent parameters and action space that the intelligent agent is responsible for regulating. It employs the Deep Deterministic Policy Gradient (DDPG) algorithm to update the agent parameters, including: Based on the Multi-Agent Reinforcement Learning (MARL) algorithm, the agent parameters that the intelligent agent is responsible for regulating are defined, including controlling microwave power, controlling gas flow rate, and controlling SF6 concentration stability. The state space of the agent parameters is defined, including microwave power, temperature, concentration, and gas flow rate. The action space is defined, including the adjustment amount of microwave power and the change amount of gas flow rate. The safety boundary is determined based on historical experience. The Deep Deterministic Policy Gradient (DDPG) algorithm was selected for training, and state updates were performed based on the agent parameters, with rewards calculated based on the concentration error.
5. The intelligent regulation method for SF6 waste gas microwave plasma treatment of claim 1, wherein, The method employs a centralized commentator algorithm to calculate the temporal difference error and iteratively update the action space parameters, including: The agent's policy network is input with agent parameters and action space, and the neural ODE model is input to determine SF6 concentration and temperature data. Rewards are calculated in a timely manner. A centralized critic algorithm is used to calculate the temporal difference error by forming quadruples based on agent parameters and action space parameters. Gradient descent calculation is performed with the goal of minimizing the mean square error. The surrogate parameters and action space parameters are updated. When the average concentration error is less than or equal to the convergence threshold, the update of the output action space parameters is stopped as the optimal control strategy.
6. The intelligent regulation method for SF6 waste gas microwave plasma treatment of claim 1, wherein, The process of saving data and generating log data for cloud storage includes: The output action space parameters and agent parameters are saved, logs are generated using the ELK Stack tool, encrypted with SSL / TLS communication, and stored in the cloud.
7. The intelligent regulation method for SF6 waste gas microwave plasma treatment of claim 1, wherein, The division of the SF6 exhaust gas treatment chamber into zones includes: The microwave treatment cavity for SF6 waste gas treatment is divided into zones, including a reaction chamber, a gas inlet, and a gas outlet.
8. An intelligent control system suitable for SF6 waste gas microwave plasma treatment, based on the intelligent control method suitable for SF6 waste gas microwave plasma treatment according to any one of claims 1-7, characterized in that, include: The equilibrium modeling module is used to establish and discretely update particle and energy balance equations according to cavity regions. The fading modeling module is used to calculate the spatial distribution of field strength based on the combined attenuation characterization of geometric distance and free space path loss superimposed with stochastic small-scale effects. The beamforming receiving signal module is used to couple multi-source power distribution with integrated attenuation and output the received signal strength at each measurement point; The external source power estimation module is used to calibrate the proportionality constant using historical data to achieve online mapping from received strength to external source power; The energy model module is used to model and discretely solve the temperature based on energy conservation and heat loss, providing the time-series changes of temperature at multiple measurement points. The Neural ODE Dynamics module is used to integrate temporal features such as concentration, temperature, and external power to construct state differential equations with physical priors. The rate assessment module is used to combine inflow and outflow rates with concentration field to calculate convection term and net reaction rate, characterizing mass transfer and reaction coupling effect. The multi-agent module is used to define the state, action space, and safety boundary, and to make decisions on power and flow rate respectively to coordinate and stabilize the concentration. The reward assessment module is used to comprehensively consider concentration error, energy consumption, and temperature / concentration safety penalties. The data and log subsystem is used to persist critical states and control variables, generate searchable logs, and securely upload them to the cloud.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the intelligent control method for microwave plasma treatment of SF6 waste gas as described in any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the intelligent control method for microwave plasma treatment of SF6 waste gas as described in any one of claims 1 to 7.