An electromagnetic cracking reactor intelligent regulation and control system and method based on digital twinning

Abstract

The application discloses an electromagnetic cracking reactor intelligent regulation and control system and method based on digital twinning, and relates to the technical field of intelligent control and organic solid waste resource utilization.The system comprises a physical reactor entity, a data acquisition and processing module, a multi-physical field full-coupling simulation model, an intelligent optimization module, a closed-loop control execution module, and a fault diagnosis and prediction module.The method comprises data acquisition, construction of a full-coupling simulation model to realize high-fidelity real-time mapping, use of a reinforcement learning algorithm to perform virtual iterative training in a digital twinning environment to generate an optimal control strategy, issuance of the optimal control strategy to an execution mechanism via a closed-loop control execution module, and realization of predictive maintenance based on virtual-real residual analysis.According to the above technical scheme, the application can realize accurate temperature control, reduce product yield fluctuations, provide early warning of coking and faults, and reduce system energy consumption, thereby providing an accurate, self-adaptive and predictable intelligent regulation and control technical scheme for organic solid waste resource utilization.

Description

Technical Field

[0001] This invention relates to the field of intelligent control and organic solid waste resource utilization technology, and more specifically, to an intelligent control system and method for an electromagnetic pyrolysis reactor based on digital twins. Background Technology

[0002] Electromagnetic pyrolysis technology, as a highly efficient means of resource recovery for organic solid waste, involves strongly coupled nonlinear interactions between electromagnetic fields, temperature fields, flow fields, and chemical reaction kinetics, placing extremely high demands on the accuracy, real-time performance, and adaptability of the control system. With the deep penetration of intelligent sensing technology, digital twin technology, and reinforcement learning algorithms into the field of industrial control, the control and operation of electromagnetic pyrolysis reactors are entering a technological transformation window from passive regulation to proactive optimization.

[0003] However, existing control schemes for electromagnetic pyrolysis reactors still have the following technical shortcomings: First, there is a mismatch between control lag and nonlinearity: electromagnetic pyrolysis involves a strongly coupled nonlinear process of "electro-thermal-chemical reaction". Traditional PID relies on manual tuning, which is difficult to cope with the large lag characteristics, resulting in a temperature control accuracy of only ±5 to 10℃. Secondly, the black box is uncontrollable: coking and temperature distribution inside the reactor cannot be detected in real time, resulting in product yield fluctuations of up to ±10% to 15%. Third, maintenance is passive: fault diagnosis is mostly based on threshold alarms, which lacks predictability and often leads to unplanned downtime.

[0004] In summary, to address the shortcomings of the existing technologies, this invention provides an intelligent control system and method for electromagnetic pyrolysis reactors based on digital twins, aiming to achieve precise temperature control, early fault warning, and multi-objective energy consumption optimization. Summary of the Invention

[0005] The main objective of this invention is to disclose an intelligent control system for an electromagnetic pyrolysis reactor based on digital twins, comprising: Data acquisition and processing module: used to acquire operational data collected from the physical reactor entity, and to clean, denoise, extract features, and deploy edge computing on the operational data to output structured feature data; the operational data includes: temperature, pressure, flow rate, gas composition, and electromagnetic power; Specifically, the data acquisition and processing module is deployed on the edge gateway, and sequentially performs abnormal data cleaning based on physical rational domain and median filtering, noise suppression based on wavelet or Kalman filtering, and time-domain statistics and frequency-domain feature extraction on the edge side.

[0006] Multiphysics fully coupled simulation model: used to solve the structured feature data through multiphysics coupling, and output the predicted results of temperature field distribution, product ratio, coking trend and energy consumption; the multiphysics fully coupled simulation model supports multiphysics fields such as electromagnetic field, temperature field, flow field and reaction dynamics; Specifically, the multiphysics fully coupled simulation model includes: Electromagnetic field sub-model: Based on Maxwell's equations, the electromagnetic induction heating process is solved, and the electromagnetic induced heat source generated by induction heating and local hot spots in the absorbing medium is calculated. This serves as the heat source input for the temperature field sub-model and the electromagnetic force input for the fluid dynamics of the flow field sub-model. Temperature field sub-model: Based on the heat transfer equation, the temperature distribution inside the reactor is solved, and the bulk control temperature of the reactor and the local reaction hot spot temperature on the surface of the absorbing medium are distinguished. Flow field sub-model: Solving the material flow and mixing characteristics within the reactor based on fluid dynamics equations; Reaction kinetics sub-model: Based on the chemical reaction kinetic equation, the cracking reaction process is solved, and the calculated results of temperature field and flow field are used as boundary conditions to output the predicted results.

[0007] Intelligent optimization module: Based on the prediction results, a multi-objective reward function is constructed using a reinforcement learning algorithm, and an optimal control strategy is generated through virtual iterative training; the optimal control strategy includes: optimized parameters for target temperature, feed rate, and electromagnetic power; Specifically, the intelligent optimization module performs virtual iterative training through state observation, action output, and multi-objective reward function scoring until convergence; the state space refers to the state of observed temperature, pressure, gas composition, vibration, displacement, and derived characteristics such as temperature gradient, heating rate, and product yield; the action space refers to the continuous and discrete control quantities that adjust electromagnetic field strength, feed rate, stirring, ball chain speed, and carbon powder circulation ratio as action outputs.

[0008] Closed-loop control execution module: used to map the optimal control strategy to the set value of the actuator, send it to the physical reactor entity through the data acquisition and processing module, and receive the latest operating data collected by the physical reactor entity executing the optimal control strategy, generate monitoring operating data and send it to the multiphysics fully coupled simulation model for state update and model correction, thus forming closed-loop control.

[0009] Fault diagnosis and prediction module: It is used to calculate the virtual and real residuals based on the running data and the prediction results of the multiphysics fully coupled simulation model, and compare them with the preset threshold. When the virtual and real residuals exceed the preset threshold, a fault warning is triggered to realize predictive maintenance. Specifically, the preset threshold includes two layers: preferred default configuration and determination method; the default configuration refers to the initial threshold selected by simulation verification and engineering experience; the health condition statistical method refers to: collecting steady-state data under historical normal operating conditions and calculating the mean and standard deviation of the normalized residuals of each monitoring variable; the accuracy fusion method refers to: the sensor accuracy level based on the physical reactor entity and the prediction uncertainty of the multi-physics fully coupled simulation model.

[0010] Furthermore, the fault diagnosis and prediction module generates maintenance strategies based on the fault type; the fault types include: malfunction, coking, sensor drift, uneven heating, and abnormal feeding; the maintenance strategies include: adjusting the electromagnetic field distribution to suppress coking.

[0011] On the other hand, the present invention provides an intelligent control method for an electromagnetic pyrolysis reactor based on digital twins, comprising: The operation data collected from the physical reactor are obtained; the operation data includes: temperature, pressure, flow rate, gas composition and electromagnetic power.

[0012] The operational data is cleaned, denoised, feature-extracted, and edge computing deployed to obtain structured feature data.

[0013] Based on the structured feature data, a multi-physics field fully coupled simulation model is constructed through multi-physics field coupling solution, and the predicted results of temperature field distribution, product ratio, coking trend and energy consumption are output.

[0014] Based on the prediction results, a multi-objective reward function is constructed using a reinforcement learning algorithm, and an optimal control strategy is generated through virtual iterative training; the optimal control strategy includes optimized parameters for target temperature, feed rate, and electromagnetic power; Specifically, the training environment for the virtual iterative training is: a multi-physics fully coupled simulation model serves as an environment simulator for reinforcement learning, replacing real experiments, and is used to achieve safe exploration and tens of thousands of virtual iterations; Furthermore, the multi-objective reward function is composed of a weighted sum of the normalized values ​​of product yield, unit energy consumption, and coking index; the weight coefficient of the product yield is greater than the weight coefficients of unit energy consumption and coking index.

[0015] The optimal control strategy is mapped to the set value of the actuator, and sent to the physical reactor entity through the data acquisition and processing module. The module also receives the latest operating data collected by the physical reactor entity executing the optimal control strategy, generates monitoring operating data, and sends it to the multiphysics fully coupled simulation model for state update and model correction, thus forming a closed-loop control. Specifically, the closed-loop control refers to feeding back the monitoring and operation data to the multiphysics fully coupled simulation model for rolling correction of the multiphysics fully coupled simulation model.

[0016] Based on the operational data and prediction results, the virtual and real residuals are calculated and compared with a preset threshold. When the virtual and real residuals exceed the preset threshold, a fault warning is triggered to achieve predictive maintenance.

[0017] According to this invention, a high-fidelity real-time mapping of the entire electromagnetic-thermal-fluidic-chemical process is performed through a multi-physics fully coupled simulation model. Combined with real-time feature extraction from the edge side of the data acquisition and processing module, a reinforcement learning algorithm is used to construct a multi-objective reward function for virtual iterative training to generate the optimal control strategy. At the same time, based on the virtual-real residual analysis of the physical reactor's measured data and simulation prediction data, fault early warning and predictive maintenance are achieved. The simulation model and algorithm verification show that temperature precision control, reduced product yield fluctuations, reduced energy consumption, and early warning of coking and faults can be achieved, improving the intelligent control level and long-term operational stability of the electromagnetic pyrolysis reactor under extreme conditions. Attached Figure Description

[0018] Figure 1 This is a system architecture diagram of an intelligent control system and method for an electromagnetic pyrolysis reactor based on digital twins, provided according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the method flow of the intelligent control system and method for an electromagnetic pyrolysis reactor based on digital twins provided in an embodiment of the present invention; Figure 3 This is a flowchart of the four-field iterative solution of the multi-physics field fully coupled simulation model of the intelligent control system and method for electromagnetic pyrolysis reactor based on digital twins, provided in an embodiment of the present invention. Detailed Implementation

[0019] The specific embodiments provided by the present invention will now be described in detail with reference to the accompanying drawings.

[0020] like Figure 1 As shown, the present invention discloses an intelligent control system for an electromagnetic pyrolysis reactor based on digital twins, comprising: P100 Data Acquisition and Processing Module: Used to acquire operational data collected from the physical reactor entity, and to clean, reduce noise, extract features, and deploy edge computing on the operational data, outputting structured feature data; The operational data includes: temperature, pressure, flow rate, gas composition, and electromagnetic power; Specifically, the data acquisition and processing module is deployed at the edge gateway. At the edge, it sequentially performs outlier cleaning based on the physical rational domain and median filtering, noise suppression based on wavelet or Kalman filtering, and time-domain statistics and frequency-domain feature extraction. The specific steps are as follows: Step 1: Cleaning: Remove frames with abnormal jumps and sensor failures (based on the physical reasonable domain and median filtering); Step 2: Noise Reduction: Use wavelet or Kalman filtering to smooth high-frequency noise; Step 3: Feature extraction: Extract time-domain statistical features (mean / variance / slope), frequency-domain features (vibration spectrum), and derived features (such as temperature gradient and heating rate) to output structured data.

[0021] P110 Multiphysics Fully Coupled Simulation Model: Used to solve the structured feature data through multiphysics coupling, outputting prediction results of temperature field distribution, product ratio, coking trend and energy consumption; the multiphysics fully coupled simulation model supports multiphysics fields such as electromagnetic field, temperature field, flow field and reaction dynamics; Among them, the multiphysics fully coupled simulation model uses P100 structured feature data as the driving boundary conditions and electromagnetically induced heat source distribution (including local hot spots of the absorbing medium) generated by electromagnetic field solution as the heat source input of temperature field. Temperature field and flow field jointly determine the heat and mass transfer characteristics in reactor and serve as boundary constraints for reaction dynamics. High-fidelity simulation is achieved through sequential coupling iterative solution and is synchronized with real-time data driving of physical reactor entity. like Figure 3 As shown, the multiphysics fully coupled simulation model includes: P310 Electromagnetic Field Sub-model: Based on Maxwell's equations, the electromagnetic induction heating process is solved, and the electromagnetic induced heat source generated by induction heating and local hot spots in the absorbing medium is calculated. This serves as the heat source input for the temperature field sub-model and the electromagnetic force input for the fluid dynamics of the flow field sub-model. The electromagnetic field sub-model is solved using Maxwell's equations in the frequency domain harmonic state to obtain the spatially distributed electromagnetically induced heat source term, the specific formula of which is as follows: ×E=-jωμH, ×H=(σ+jωε)E, ·(εE)=ρ_v, ·(μH)=0; Where E is the electric field strength (V / m), H is the magnetic field strength (A / m), σ is the electrical conductivity (S / m), and ε = ε0(ε r ′-jε r ") is the complex permittivity, μ = μ0(μ r ′-jμ r ″) is the complex permeability, ω=2πf is the angular frequency (f is the operating frequency, 5kHz for external wave sources and 1–2kHz for internal wave sources); Specifically, the volumetric heat generation rate (i.e., the electromagnetically induced heat source term) is composed of three coupled parts: eddy current / Joule heating, dielectric loss, and magnetic loss. The heat source is mainly generated locally on the surface and inside the absorbing medium (circulating carbon powder, loose nickel balls), serving as the non-uniform spatial heat source input for the temperature field sub-model.

[0022] P320 Temperature Field Sub-model: Based on the heat transfer equation, the temperature distribution inside the reactor is solved, and the bulk control temperature of the reactor and the local reaction hot spot temperature on the surface of the absorbing medium are distinguished. Specifically, the temperature field sub-model has a closed-loop control accuracy of ±1 to 2℃ for the bulk temperature. The local hot spots (approximately 500 to 700℃) on the surface of the absorbing medium (such as carbon-based media, nickel spheres, etc.) are determined by the absorption characteristics of the medium and are decoupled from the bulk temperature. The electromagnetically induced heat source (including local hot spots in the absorbing medium) output by the electromagnetic field sub-model is used as the heat source input; the temperature field distribution data output by the temperature field sub-model is simultaneously input to the flow field sub-model and the reaction dynamics sub-model as boundary conditions for flow field driving and reaction rate calculation. The temperature field sub-model uses q_EM(r) as the heat source term to solve the transient energy conservation equation, and the specific formula is as follows: ρc_p( T / t+u· T)= ·(k T)+q_EM(r)+q_rxn; Where ρ is the density (kg / m³) 3 c_p is the specific heat capacity (J / (kg·K)), u is the velocity field (m / s), k is the effective thermal conductivity (W / (m·K)), and q_rxn=-ΔH_rxn·Σr_i is the heat of reaction (W / m). 3 ), r_i is the reaction rate of the i-th lumped reaction (kg / (m²)). 3 ·s)); Specifically, in order to characterize the dual-scale features of "low temperature in the bulk phase and high temperature hot spot on the surface of the absorbing medium" in this invention, the temperature field is decoupled into the bulk control temperature T_b and the local hot spot temperature T_s on the surface of the absorbing medium. For example, at a bulk control temperature T_b≈350℃, the hot spot temperature T_s on the surface of the absorbing medium can reach 500-700℃; achieving closed-loop control precision of the bulk control temperature T_b, the local hot spot temperature of the reaction site is determined by the absorption characteristics of the medium and decoupled from the bulk temperature.

[0023] P330 flow field sub-model: Solving the material flow and mixing characteristics in the reactor based on fluid dynamics equations; The flow field sub-model uses the Euler-Euler multiphase / porous medium model to solve the mass conservation and momentum equations, characterizing the flow and mixing of materials and loose nickel balls in the stirred bed.

[0024] P340 reaction kinetics sub-model: Based on the chemical reaction kinetic equation, the cracking reaction process is solved, and the calculated results of temperature field and flow field are used as boundary conditions to output product distribution prediction and coking trend prediction. Specifically, the reaction kinetics sub-model adopts a lumped component-Arrhenius coupled model, with the hot spot temperature T_s on the surface of the absorbing medium given by the temperature field sub-model and the residence time given by the flow field sub-model as boundary conditions.

[0025] P120 Intelligent Optimization Module: Based on the prediction results, a multi-objective reward function is constructed using a reinforcement learning algorithm, and an optimal control strategy is generated through virtual iterative training; the optimal control strategy includes: optimized parameters for target temperature, feed rate, and electromagnetic power; Specifically, the intelligent optimization module uses a multi-physics fully coupled simulation model as the training environment, the prediction results output by P110 as the state input, and electromagnetic field strength, feed rate, and stirring speed as the action space output. Virtual iterative training is performed through PPO until the policy network converges to the optimal control policy. The specific steps are as follows: 1) State space S (state observation): temperature (each segment), pressure, gas composition, vibration / displacement, and derived features (temperature gradient, heating rate, product yield estimation).

[0026] 2) Action Space A (Action Output): Adjusts continuous or discrete control quantities such as electromagnetic field strength (coil power / frequency), feed rate, stirring / ball chain speed, and carbon powder circulation ratio.

[0027] 3) Multi-objective reward function R: The multi-objective reward function is composed of the weighted sum of the normalized values ​​of product yield, unit energy consumption and coking index, wherein the weight coefficient of product yield is greater than the weight coefficients of energy consumption penalty term and coking penalty term. The specific formulas for summing the multi-objective reward functions and the weighting coefficients are as follows: R=w_y·Y_n-w_e·E_n-w_c·C_n; Wherein, Y_n is the normalized product yield, E_n is the normalized unit energy consumption (penalty term), C_n is the normalized coking index (penalty term), and w_y, w_e, and w_c are the weight coefficients of the three, with the following weight coefficients: w_y=0.55, w_e=0.30, w_c=0.15 (the sum of the three = 1.00), that is, product yield is the primary factor, followed by energy consumption and coking; the weights are the optimal configuration for simulation verification, and can be optimized online in the [0,1] interval according to the operating conditions and target preferences.

[0028] P130 Closed-Loop Control Execution Module: This module maps the optimal control strategy to the setpoint of the actuator, sends the data to the physical reactor entity through the data acquisition and processing module, receives the latest operating data collected by the physical reactor entity executing the optimal control strategy, generates monitoring operating data, and sends it to the multiphysics fully coupled simulation model for state updates and model correction, thus forming a closed-loop control. The monitoring and operation data include: real-time monitoring data of temperature, pressure, flow rate, gas composition and electromagnetic power (including heating power supply current and voltage), as well as product yield, temperature gradient and heating rate determined based on the above monitoring data. Specifically, the closed-loop control execution module supports Modbus / TCP or OPC UA industrial communication protocols, with a control cycle of 0.5s to 1s; The optimized control parameters are converted into analog or digital signals by the I / O interface module in the PLC / DCS control cabinet, which respectively drive the electromagnetic heating power supply to adjust the output power, drive the screw feed valve to adjust the opening, and drive the stirring motor to adjust the speed.

[0029] P140 Fault Diagnosis and Prediction Module: Used to acquire the running data of the data acquisition and processing module and the prediction results of the multiphysics fully coupled simulation model, calculate the virtual and real residuals and compare them with a preset threshold, trigger a fault warning when the residuals exceed the preset threshold, and realize predictive maintenance; Specifically, the fault diagnosis and prediction module synchronously acquires the running data obtained by P100 and the prediction results output by P110, calculates the deviation between the measured data and the simulation prediction data at the same time, and generates a virtual and real residual vector. The method for calculating the virtual and real residual vectors is as follows: Define an m-dimensional state vector synchronously sampled at time t, where each component corresponds to the same set of monitored process variables (temperature, pressure, flow rate, and gas composition H2, CH4, CO, CO2). The measured vector is denoted as y_meas(t)∈R^m, and the predicted vector is denoted as y_sim(t)∈R^m. The two are aligned according to the control period (0.5s~1s). The specific formula is as follows: Virtual and real residual vector: r(t) = y_meas(t) - y_sim(t) = [r_1(t), r_2(t), ..., r_m(t)]^T; Wherein, the i-th dimension component is r_i(t)=y_i,meas(t)-y_i,sim(t), i=1,2,…,m; Specifically, to eliminate dimensional differences and facilitate cross-dimensional comparisons, a normalized residual is defined, with the following formula: r_i,n(t)=|y_i,meas(t)-y_i,sim(t)| / Y_i,ref; Where Y_i,ref is the nominal / standard reference value of the i-th dimension variable.

[0030] Each dimension component in the virtual and real residual vector is compared with the corresponding preset threshold: when the absolute value of all virtual and real residual components r_i,n(t) is less than the corresponding preset threshold, it is determined to be normal operation; when the virtual and real residual is greater than or equal to the preset threshold, it is determined to be coking or equipment failure and an early warning signal is triggered; when the global composite virtual and real residual R_n(t) is greater than or equal to the preset threshold, an early warning is also triggered. The specific formula for the global composite virtual-real residual R_n(t) is as follows: R_n(t)=sqrt(Σ_i(r_i,n(t))^2); The preset threshold setting includes two layers: a preferred default configuration and a determination method (health condition statistical method and accuracy fusion method), specifically: 1) Optimize the default configuration (the initial threshold selected based on simulation verification and engineering experience, to be fine-tuned after calibration with measured data; normalized dimensionless values, with Y_i,ref as a reference, can be optimized online within the [0,1] interval): Temperature (bulk phase T_b): ε_T = 1% (approximately equivalent to ±3.5℃ (1% ≈ 3.5℃ at 350℃ under rated conditions), above ±1–2℃ control band to distinguish sensor noise from actual anomalies); Pressure: ε_P = 3%; Flow rate: ε_Q = 3%; Energy consumption: ε_E = 3%; Gas composition (H2, CH4, CO, CO2 components): ε_comp = 5%; Coking trend: ε_coke = 8%; Global cohesion: ε_g = 6%.

[0031] 2) The health condition statistical method and the accuracy fusion method require measured data to calculate specific thresholds. In the initial stage of system operation, the system starts with the preferred default configuration. Once all measured data is available, the system is recalculated and fine-tuned. The specific method is as follows: Health condition statistical method: Collect N sets of steady-state data under normal operating conditions with no coking or faults in history, and calculate the mean μ_i and standard deviation σ_i of the i-th dimension normalized residual. The specific formula is as follows: ε_i = μ_i + k·σ_i; Where k=3 (3σ principle, covering approximately 99.7% of normal fluctuations); Accuracy fusion method: Based on the sensor accuracy level and the multi-physics fully coupled simulation model, the uncertainty is predicted. The specific formula is as follows: ε_i=S·sqrt(δ_s,i^2+δ_m,i^2); Wherein, the safety factor S = 2~3, δ_m is the uncertainty predicted by the multiphysics fully coupled simulation model, and δ_s is the sensor accuracy level.

[0032] When the problem is determined to be coking or equipment failure, a maintenance strategy is generated based on the type of failure. The maintenance strategy includes adjusting the electromagnetic field distribution to suppress coking.

[0033] This invention also provides a digital twin-based intelligent control method for electromagnetic pyrolysis reactors, which is implemented using the aforementioned digital twin-based intelligent control system for electromagnetic pyrolysis reactors. like Figure 2 As shown, the intelligent control method for an electromagnetic pyrolysis reactor based on digital twins includes: Step S200: Obtain operational data collected from the physical reactor entity; The operational data includes: temperature, pressure, flow rate, gas composition, and electromagnetic power.

[0034] Step S210: Clean, denoise, extract features, and deploy edge computing on the running data to obtain structured feature data; The data acquisition and processing module is deployed at the edge gateway. At the edge, it sequentially performs outlier cleaning based on the physical rational domain and median filtering, noise suppression based on wavelet or Kalman filtering, and time-domain statistics and frequency-domain feature extraction. The specific steps are as follows: Step 1: Cleaning – Removing frames with abnormal jumps and sensor failures, based on the physical reasonable domain and median filtering; Step 2: Noise Reduction – Use wavelet or Kalman filtering to smooth high-frequency noise; Step 3: Feature Extraction – Extract time-domain statistical features (mean, variance, slope), frequency-domain features (vibration spectrum), and derived features (such as temperature gradient and heating rate) to output structured feature data.

[0035] Step S220: Based on the structured feature data, a multi-physics field fully coupled simulation model is constructed through multi-physics field coupling solution, and the predicted results of temperature field distribution, product ratio, coking trend and energy consumption are output. Specifically, the multiphysics fully coupled simulation model uses the structured feature data as the driving boundary conditions and the electromagnetically induced heat source distribution (including local hot spots in the absorbing medium) generated by solving the electromagnetic field as the heat source input of the temperature field. The temperature field and the flow field jointly determine the heat and mass transfer characteristics in the reactor and serve as the boundary constraints of the reaction dynamics. High-fidelity simulation is achieved through sequential coupling and iterative solving of the four fields, and is synchronized with the real-time data driving of the physical reactor entity.

[0036] The multiphysics fully coupled simulation model includes an electromagnetic field sub-model, a temperature field sub-model, a flow field sub-model, and a reaction dynamics sub-model. Specifically, the electromagnetic field sub-model obtains the electromagnetically induced heat source by solving Maxwell's equations, which serves as the heat source input for the temperature field sub-model. The temperature field sub-model solves the transient energy conservation equation and outputs the temperature field distribution, which is then input into the flow field sub-model and the reaction kinetics sub-model. The flow field sub-model solves the mass conservation and momentum equations, and outputs the material flow and mixing characteristics. The reaction kinetics sub-model solves the cracking reaction process based on the chemical reaction kinetic equation, and outputs predictions of product distribution and coking trends.

[0037] Step S230: Based on the prediction results, a multi-objective reward function is constructed using a reinforcement learning algorithm, and an optimal control strategy is generated through virtual iterative training; the optimal control strategy includes optimized parameters for target temperature, feed rate, and electromagnetic power; Specifically, the current state is obtained from the multiphysics fully coupled simulation model, and the current state is input into the policy network to infer and output action parameters; the action parameters are sent to the multiphysics fully coupled simulation model to perform a simulation step-long inference to obtain the next state and the immediate reward value; the state, action, reward value and the next state are stored in the experience replay buffer, and the PPO algorithm is used to update the gradient of the policy network. The process is repeated until the policy network converges. The multi-objective reward function is composed of a weighted sum of the normalized values ​​of product yield, unit energy consumption, and coking index, wherein the weight coefficient of product yield is greater than the weight coefficients of energy consumption penalty term and coking penalty term.

[0038] Step S240: The optimal control strategy is mapped to the set value of the actuator, sent to the physical reactor entity through the data acquisition and processing module, and the latest operating data collected by the physical reactor entity executing the optimal control strategy is received. Monitoring operating data is generated and sent to the multiphysics fully coupled simulation model for state update and model correction to form closed-loop control. Specifically, the closed-loop control execution module supports Modbus / TCP or OPC UA industrial communication protocols, with a control cycle of 0.5s to 1s. The optimized control parameters are converted into analog or digital signals by the I / O interface module in the PLC / DCS control cabinet, which respectively drive the electromagnetic heating power supply to adjust the output power, drive the screw feed valve to adjust the opening, and drive the stirring motor to adjust the speed. The monitoring and operation data include: real-time monitoring data of temperature, pressure, flow rate, gas composition and electromagnetic power (including heating power supply current and voltage), as well as product yield, temperature gradient and heating rate determined based on the above monitoring data.

[0039] In this implementation step, the closed-loop control refers to feeding back the monitoring operation data to the multiphysics fully coupled simulation model for rolling correction of the multiphysics fully coupled simulation model.

[0040] Step S250: Based on the running data and prediction results, calculate the virtual and real residuals and compare them with a preset threshold. When the residuals exceed the preset threshold, trigger a fault warning to achieve predictive maintenance. Specifically, the running data of S200 and the prediction results output by S220 are acquired simultaneously, the deviation between the measured data and the simulation prediction data at the same time is calculated, and the virtual and real residual vector is generated. Each dimension component in the virtual-real residual vector is compared with the corresponding preset threshold. When the absolute value of all virtual-real residual components is less than the corresponding preset threshold, it is determined to be normal operation. When the virtual-real residual component is greater than or equal to the preset threshold, it is determined to be coking or equipment failure and an early warning signal is triggered. When the global composite virtual-real residual R_n(t) is greater than or equal to the preset threshold, an early warning is triggered.

[0041] This invention, based on the deep integration of multi-physics digital twins and reinforcement learning, constructs a fully coupled simulation model of electromagnetic, thermal, fluidic, and chemical fields, achieving high-fidelity real-time mapping of the reaction process. The PPO algorithm is used for virtual iterative training in the digital twin environment to generate the optimal control strategy. Combined with a closed-loop control execution module, it achieves millisecond-level response from virtual optimization to physical execution. Furthermore, a virtual-real residual fault diagnosis mechanism is introduced to achieve predictive maintenance, effectively solving technical problems such as control lag, uncontrollable internal state black box, and passive fault diagnosis in electromagnetic pyrolysis reactors. This results in improved temperature control accuracy, reduced product yield fluctuations, early warning of coking and faults, reduced system energy consumption, and increased automation, providing a precise, adaptive, and predictable intelligent control technology solution for the resource-based treatment of organic solid waste.

[0042] The above-disclosed embodiments are merely a few specific examples of the present invention. However, the present invention is not limited thereto, and any variations that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A digital twin-based intelligent control system for an electromagnetic pyrolysis reactor, characterized in that, include: Data acquisition and processing module: used to acquire operational data collected from the physical reactor entity, and to clean, denoise, extract features, and deploy edge computing on the operational data to output structured feature data; the operational data includes: temperature, pressure, flow rate, gas composition, and electromagnetic power; Multiphysics fully coupled simulation model: used to solve the structured feature data through multiphysics coupling, and output the predicted results of temperature field distribution, product ratio, coking trend and energy consumption; the multiphysics fully coupled simulation model supports multiphysics fields such as electromagnetic field, temperature field, flow field and reaction dynamics; Intelligent optimization module: Based on the prediction results, a multi-objective reward function is constructed using a reinforcement learning algorithm, and an optimal control strategy is generated through virtual iterative training; the optimal control strategy includes: optimized parameters for target temperature, feed rate, and electromagnetic power; Closed-loop control execution module: used to map the optimal control strategy to the set value of the actuator, send it to the physical reactor entity through the data acquisition and processing module, and receive the latest operating data collected by the physical reactor entity executing the optimal control strategy, generate monitoring operating data and send it to the multiphysics fully coupled simulation model for state update and model correction, thus forming closed-loop control; Fault diagnosis and prediction module: It is used to calculate the virtual and real residuals based on the running data and the prediction results of the multiphysics fully coupled simulation model, and compare them with the preset threshold. When the virtual and real residuals exceed the preset threshold, a fault warning is triggered to realize predictive maintenance.

2. The intelligent control system for an electromagnetic pyrolysis reactor based on digital twins according to claim 1, characterized in that, The data acquisition and processing module is deployed on the edge gateway, and sequentially performs abnormal data cleaning based on physical rational domain and median filtering, noise suppression based on wavelet or Kalman filtering, and time-domain statistics and frequency-domain feature extraction on the edge side.

3. The intelligent control system for an electromagnetic pyrolysis reactor based on digital twins according to claim 1, characterized in that, The multiphysics fully coupled simulation model includes: Electromagnetic field sub-model: Based on Maxwell's equations, the electromagnetic induction heating process is solved, and the electromagnetic induced heat source generated by induction heating and local hot spots in the absorbing medium is calculated. This serves as the heat source input for the temperature field sub-model and the electromagnetic force input for the fluid dynamics of the flow field sub-model. Temperature field sub-model: Based on the heat transfer equation, the temperature distribution inside the reactor is solved, and the bulk control temperature of the reactor and the local reaction hot spot temperature on the surface of the absorbing medium are distinguished. Flow field sub-model: Solving the material flow and mixing characteristics within the reactor based on fluid dynamics equations; Reaction kinetics sub-model: Based on the chemical reaction kinetic equation, the cracking reaction process is solved, and the calculated results of temperature field and flow field are used as boundary conditions to output the predicted results.

4. The intelligent control system for an electromagnetic pyrolysis reactor based on digital twins according to claim 1, characterized in that, The intelligent optimization module performs virtual iterative training through state observation, action output, and multi-objective reward function scoring until convergence. The state space refers to the state of observed temperature, pressure, gas composition, vibration, displacement, and derived characteristics such as temperature gradient, heating rate, and product yield. The action space refers to the continuous and discrete control quantities that adjust electromagnetic field strength, feed rate, stirring, ball chain speed, and carbon powder circulation ratio as action outputs.

5. The intelligent control system for an electromagnetic pyrolysis reactor based on digital twins according to claim 1, characterized in that, The preset threshold includes two layers: preferred default configuration and determination method; the default configuration refers to the initial threshold selected by simulation verification and engineering experience; the health condition statistical method refers to: collecting steady-state data under historical normal operating conditions and calculating the mean and standard deviation of the normalized residuals of each monitoring variable; the accuracy fusion method refers to: the sensor accuracy level based on the physical reactor entity and the prediction uncertainty of the multi-physics field fully coupled simulation model.

6. The intelligent control system for an electromagnetic pyrolysis reactor based on digital twins according to claim 1, characterized in that, The fault diagnosis and prediction module generates maintenance strategies based on the fault type. The fault types include: malfunction, coking, sensor drift, uneven heating, and abnormal feeding; the maintenance strategies include: adjusting the electromagnetic field distribution to suppress coking.

7. A method for intelligent control of an electromagnetic pyrolysis reactor based on digital twins, characterized in that, The system is implemented using the intelligent control system for an electromagnetic pyrolysis reactor based on digital twins as described in any one of claims 1-6, comprising: Acquire operational data collected from the physical reactor entity; the operational data includes: temperature, pressure, flow rate, gas composition, and electromagnetic power; The operational data is cleaned, denoised, feature-extracted, and edge computing deployed to obtain structured feature data; Based on the structured feature data, a multi-physics field fully coupled simulation model is constructed through multi-physics field coupling solution, and the predicted results of temperature field distribution, product ratio, coking trend and energy consumption are output. Based on the prediction results, a multi-objective reward function is constructed using a reinforcement learning algorithm, and an optimal control strategy is generated through virtual iterative training; the optimal control strategy includes optimized parameters for target temperature, feed rate, and electromagnetic power; The optimal control strategy is mapped to the set value of the actuator, and sent to the physical reactor entity through the data acquisition and processing module. The module also receives the latest operating data collected by the physical reactor entity executing the optimal control strategy, generates monitoring operating data, and sends it to the multiphysics fully coupled simulation model for state update and model correction, thus forming a closed-loop control. Based on the operational data and prediction results, the virtual and real residuals are calculated and compared with a preset threshold. When the virtual and real residuals exceed the preset threshold, a fault warning is triggered to achieve predictive maintenance.

8. The intelligent control method for an electromagnetic pyrolysis reactor based on digital twins according to claim 7, characterized in that, The training environment for the virtual iterative training is a multi-physics fully coupled simulation model, which serves as an environment simulator for reinforcement learning, replacing real experiments and enabling safe exploration and tens of thousands of virtual iterations.

9. The intelligent control method for an electromagnetic pyrolysis reactor based on digital twins according to claim 7, characterized in that, The multi-objective reward function is composed of a weighted sum of the normalized values ​​of product yield, unit energy consumption, and coking index; the weight coefficient of product yield is greater than the weight coefficients of unit energy consumption and coking index.

10. The intelligent control method for an electromagnetic pyrolysis reactor based on digital twins according to claim 7, characterized in that, The closed-loop control refers to feeding back the monitoring and operation data to the multiphysics fully coupled simulation model for rolling correction of the multiphysics fully coupled simulation model.

Images