A photovoltaic module lamination process parameter collaborative optimization system based on process digital twin

Abstract

This invention relates to a collaborative optimization system for photovoltaic module stacking process parameters based on a digital twin of the process, comprising: a multiphysics coupled simulation module, which constructs a high-fidelity digital twin through multiphysics coupled simulation, coupling electromagnetic transmission, heat conduction, and fluid dynamics equations, and employing adaptive finite element discretization technology to achieve efficient and high-precision simulation; a meta-learning optimization module, based on the MAML-PINN architecture, which quickly masters new process rules through a small batch of experiments; and a process disturbance self-correction module, which compensates for environmental temperature drift and equipment aging in real time, and improves the success rate of parameter adaptation across production lines through knowledge transfer, and locates the causes of defects. These three modules form a closed-loop optimization through data-model-control three-flow coupling, dynamically adjusting key parameters such as wafer spacing and welding temperature. This invention significantly improves the simulation efficiency, parameter adaptability, and disturbance resistance of photovoltaic module stacking processes, thereby increasing production efficiency and product yield.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic module manufacturing technology, and more specifically to a collaborative optimization system for photovoltaic module stacking process parameters based on a process digital twin. Background Technology

[0002] In recent years, with the continuous growth of global demand for renewable energy, the photovoltaic industry has developed rapidly, and the manufacturing process of photovoltaic modules is facing higher efficiency and quality requirements. As a key step in photovoltaic module production, the parameter optimization of the shingling process directly affects the module's performance and yield. However, traditional shingling process parameter optimization mainly relies on trial and error and offline simulation, which suffers from problems such as long simulation time, poor parameter adaptability, and difficulty in handling dynamic disturbances. For example, the causes of defects such as lamination bubbles and poor soldering are complex, involving the coupling effects of multiple physical fields such as electromagnetics, heat, and fluids, making it difficult for traditional methods to accurately trace the source and adjust in real time. Furthermore, environmental temperature drift and equipment aging further increase the challenges to process stability.

[0003] Digital twin technology offers a new approach to solving the aforementioned problems. By constructing a high-fidelity digital twin of a process, real-time simulation of multiphysics and dynamic optimization of process parameters can be achieved. The introduction of advanced algorithms such as Physical Information Neural Networks (PINN) and meta-learning can significantly improve simulation efficiency and parameter adaptation capabilities. For example, PINN reduces simulation time from hours to seconds by coupling electromagnetic transport equations, heat conduction equations, and fluid dynamics equations, combined with adaptive finite element discretization technology; meta-learning, on the other hand, quickly masters new process rules through a small number of batch experiments, achieving knowledge transfer across production lines. However, existing technologies still have shortcomings in multiphysics collaborative simulation, real-time disturbance compensation, and intelligent defect tracing, urgently requiring an integrated system to achieve closed-loop optimization of process parameters. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a photovoltaic module stacking process parameter collaborative optimization system based on a process digital twin. This system integrates multi-physics field coupling simulation, meta-learning optimization and process disturbance self-correction technology to construct a high-fidelity digital twin, realize the dynamic adjustment of process parameters and the accurate location of defect causes, thereby improving production efficiency, resource utilization and product quality.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a photovoltaic module stacking process parameter collaborative optimization system based on process digital twin, comprising:

[0006] The multiphysics coupling simulation module is used to construct a high-fidelity digital twin. By coupling electromagnetic transport equations, heat conduction equations and fluid dynamics equations, and using adaptive finite element discretization technology, it achieves efficient and high-precision simulation of photovoltaic module stacking process and outputs simulation results.

[0007] The meta-learning optimization module is connected to the multiphysics coupling simulation module. It is used to receive the simulation results and, based on the meta-learning algorithm, learn and master new process rules through a small batch of experimental data, and output the optimized process rules.

[0008] The process disturbance self-correction module is connected to the meta-learning optimization module and the multi-physics field coupled simulation module, respectively. It is used to receive the optimized process rules, and combine them with real-time production data to compensate for at least one disturbance factor in environmental temperature drift and equipment aging in real time, and to locate the cause of defects through process knowledge transfer and defect source analysis.

[0009] The multiphysics coupled simulation module, meta-learning optimization module, and process disturbance self-correction module form a closed-loop optimization through a data-model-control three-flow coupling mechanism to dynamically adjust process parameters.

[0010] Furthermore, the multiphysics coupling simulation module is further configured as follows:

[0011] Collect real-time data during the actual production process, including electric field distribution, temperature gradient, slurry flow behavior, equipment operating parameters, and defect feedback data.

[0012] The real-time data is compared with a preset threshold, and a correction process is triggered when the preset deviation is exceeded.

[0013] The data triggered during the correction process is fused with the electromagnetic transport equation, heat conduction equation, and fluid dynamics equation for simulation, and the simulation results are output to the meta-learning optimization module and the process disturbance self-correction module.

[0014] Furthermore, the meta-learning optimization module is further configured as follows:

[0015] Receive the core parameter data, including electric field strength, temperature gradient and slurry flow rate, output by the multiphysics coupling simulation module;

[0016] Using meta-learning algorithms, based on fewer than 50 sets of batch experimental data, we analyzed key process parameters including silicon wafer spacing, welding temperature, and paste concentration to master new process rules.

[0017] The optimized process rules are passed to the process disturbance self-correction module for precise compensation of environmental temperature drift and equipment aging.

[0018] Furthermore, the process disturbance self-correction module is further configured as follows:

[0019] The system receives the process rules transmitted by the meta-learning optimization module and, in conjunction with real-time production data and defect feedback data, performs real-time compensation for environmental temperature drift and equipment aging.

[0020] By utilizing process knowledge transfer technology, optimized process knowledge acquired from one production line can be applied to other production lines to improve the success rate of cross-production line parameter adaptation.

[0021] By tracing the source of defects, the probability distribution of the causes of lamination bubbles and cold solder joint defects is located, and the analysis results are fed back to the multiphysics coupling simulation module to optimize the simulation model.

[0022] Furthermore, the system is configured as follows:

[0023] When deviations in process parameters or defects in products are detected, the meta-learning optimization module and the process disturbance self-correction module adjust the process parameters in a coordinated manner based on the simulation results output by the multi-physics coupling simulation module.

[0024] Furthermore, the system employs a Model Independent Meta-Learning-Physical Information Neural Network (MAML-PINN) architecture and adaptive finite element discretization technology to achieve optimized control of process parameters, specifically including:

[0025] Step S1: Using the improved framework of the Physical Information Neural Network (PINN), the processing progress deviation is used as a constraint condition. The electromagnetic transmission equation, heat conduction equation and fluid dynamics equation are coupled to perform multi-physics coupling simulation, and the simulation results are output using adaptive finite element discretization technology.

[0026] Step S2: Based on the MAML-PINN architecture, the simulation results are processed through a meta-learning algorithm to uncover the correlation and optimization direction of process parameters and to master new process rules;

[0027] Step S3: The process disturbance self-correction module performs real-time compensation based on the learned process rules and the environmental temperature drift and equipment aging disturbance factors monitored in real time.

[0028] Further, in step S1, the electromagnetic transport equation, the heat conduction equation, and the fluid dynamics equation are specifically as follows:

[0029] Electromagnetic transport equations:

[0030] Heat conduction equation:

[0031] Fluid dynamics equations:

[0032] in, For electric field strength, For density, Where is the dielectric constant. For temperature, For time, Thermal conductivity, For specific heat capacity, The internal heat source generation rate, For fluid velocity vector, Kinematic viscosity;

[0033] The adaptive finite element discretization technique discretizes the computational domain into finite elements, and approximates the physical quantity u in spatial coordinate x as follows:

[0034]

[0035] in, Here, n represents the physical field function, and n is the number of element nodes. For shape functions, Let i be the node value at node i.

[0036] Furthermore, in step S2, the processing procedure based on the MAML-PINN architecture includes:

[0037] Suppose there are multiple tasks Each task has a corresponding dataset. For the PINN model In each task Perform an inner loop update:

[0038]

[0039] in, These are the task-specific model parameters obtained after performing inner loop learning for the j-th task. These are global, shared initial model parameters. For the inner loop learning rate, For the task The loss function;

[0040] Then, calculate the meta-loss. :

[0041]

[0042] in, To compensate for simulation error loss, For meta-learning loss, For defect-related losses, , , These are the weighting coefficients;

[0043] Then update the parameters through the outer loop. , The learning rate is the outer loop rate.

[0044] Furthermore, in step S3, the process disturbance self-correction module, based on the process rules transmitted from the meta-learning optimization module and combined with the real-time monitored environmental temperature drift, and equipment aging Perform parameter compensation;

[0045] Let the original process parameter vector be... The compensated parameter vector By compensation function Determined, that is:

[0046]

[0047] When the system detects deviations in process parameters or product defects, the meta-learning optimization module and the process disturbance self-correction module collaboratively adjust the parameters based on the simulation results output by the multiphysics coupled simulation module to minimize the total loss.

[0048] ,

[0049] in, To compensate for simulation error loss, For meta-learning loss, For defect-related losses, , , These are weighting coefficients; this ensures the stable and efficient operation of the process.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] 1. High simulation efficiency: This invention reduces the simulation time of multiphysics coupling from hours to seconds by using the improved PINN framework and adaptive finite element discretization technology.

[0052] 2. Fast parameter adaptation: Based on the meta-learning optimization module of the MAML-PINN architecture, this invention can quickly master new process rules through a small number of batch experiments, realize knowledge transfer across production lines, and improve the success rate of parameter adaptation.

[0053] 3. Strong anti-disturbance capability: In this invention, the process disturbance self-correction module can monitor and compensate for disturbances such as environmental temperature drift and equipment aging in real time, ensuring the stability of the process and the consistency of the product.

[0054] 4. Accurate Defect Source Tracing: Through closed-loop feedback and defect source tracing analysis, this invention can accurately locate the probability distribution of the causes of defects such as lamination bubbles and cold solder joints, and optimize the simulation model to continuously improve product quality. Attached Figure Description

[0055] Figure 1 This is an architecture diagram of the photovoltaic module stacking process parameter collaborative optimization system based on process digital twin provided in this embodiment of the invention;

[0056] Figure 2 This is a comparison chart of optimized process parameters in embodiments of the present invention;

[0057] Figure 3 This is a comparison chart of the real-time compensation effect of environmental temperature drift in embodiments of the present invention;

[0058] Figure 4 This is a comparison chart of the cross-production line knowledge transfer effects in an embodiment of the present invention. Detailed Implementation

[0059] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0060] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0061] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0062] like Figure 1 As shown, this embodiment provides a photovoltaic module stacking process parameter collaborative optimization system based on a process digital twin, which includes three core modules: a multiphysics coupled simulation module, a meta-learning optimization module, and a process disturbance self-correction module. The system constructs a high-fidelity digital twin using an improved Physics-Informed Neural Networks (PINN) framework, enabling rapid adaptation of process flow simulation and meta-learning-driven optimization that integrates core parameters.

[0063] The multiphysics coupling simulation module is used to construct a high-fidelity digital twin. By coupling electromagnetic transmission equations, heat conduction equations and fluid dynamics equations, and using adaptive finite element discretization technology, the simulation time is reduced from hours to seconds, realizing efficient and high-precision process simulation of photovoltaic module stacking process, and outputting simulation results.

[0064] The meta-learning optimization module is connected to the multiphysics coupling simulation module. It is used to receive the simulation results, and based on the meta-learning algorithm, learn and master new process rules through a small batch of experimental data, and output optimized process rules.

[0065] The process disturbance self-correction module is connected to both the meta-learning optimization module and the multi-physics coupled simulation module. It receives the optimized process rules and, in conjunction with real-time production data, compensates for at least one disturbance factor, such as environmental temperature drift and equipment aging, in real time. This further improves the stability and adaptability of process parameters, thereby enhancing processing accuracy. The process disturbance self-correction module also improves the success rate of cross-production line parameter (wafer spacing, welding temperature, slurry concentration, motor speed, pressure value) adaptation through process knowledge transfer and defect source analysis, and identifies the probability distribution of causes for defects such as lamination bubbles and cold solder joints.

[0066] The multiphysics coupled simulation module, meta-learning optimization module, and process disturbance self-correction module form a closed-loop optimization through a data-model-control three-flow coupling mechanism to dynamically adjust process parameters and significantly improve the efficiency and resource utilization of photovoltaic module stacking process.

[0067] In the actual production process, the system accurately collects data covering the electric field distribution between silicon wafers, the temperature gradient during lamination, the slurry flow behavior, the slurry flow rate, and the motor speed and pressure values ​​of the equipment during operation, reflecting the real-time production status. It also collects feedback data on lamination bubbles and solder joint defects. This complete and accurate data is input into the physical field coupling simulation module. The collected data is compared with a preset standard progress threshold. If the deviation exceeds a preset limit, a correction process is triggered, simultaneously collecting parameters such as the electric field distribution, temperature gradient, slurry flow data, and the motor speed and pressure values ​​of the equipment during the deviation period. This module deeply integrates the above data with electromagnetic, thermal, and fluid equations, performing complex simulation calculations on core parameters such as the electric field strength of the silicon wafers, the rate of temperature change, and the slurry flow rate. Finally, it outputs accurate and detailed simulation results to the meta-learning optimization module and the process disturbance self-correction module, laying a solid foundation of data for subsequent accurate process analysis.

[0068] The meta-learning optimization module accurately receives data from the multi-physics coupling simulation module, including core parameters such as silicon wafer electric field strength, lamination temperature gradient, and slurry flow rate. It uses advanced meta-learning algorithms to deeply analyze and process this data. With the help of a small number (less than 50 sets) of batch experimental data, it conducts research on key process parameters such as silicon wafer spacing, welding temperature, and slurry concentration to master new process rules. Subsequently, the optimized process rules involving various core parameters are passed to the process disturbance self-correction module. Based on these rules, the module accurately compensates for environmental temperature drift (±2°C) and equipment aging (±2% accuracy drift) to ensure process stability and product quality.

[0069] After receiving the process rules from the meta-learning optimization module, the process disturbance self-correction module combines real-time production data and defect feedback data to compensate for environmental temperature drift and equipment aging in real time. At the same time, it uses process knowledge transfer technology to apply the optimized process knowledge obtained from one production line to other production lines, improving the success rate of cross-production line parameter adaptation. It also uses defect source analysis to locate the probability distribution of the causes of defects such as lamination bubbles and cold welds, and then feeds the analysis results back to the multi-physics coupling simulation module to optimize the simulation model.

[0070] The simulation results output by the multiphysics coupled simulation module not only provide learning data for the meta-learning optimization module, but also provide a reference for the process adjustment of the entire system. When the system detects deviations in process parameters or defects in products, the meta-learning optimization module and the process disturbance self-correction module adjust the process parameters collaboratively based on the results of the multiphysics coupled simulation module to ensure stable and efficient operation of the process.

[0071] The system employs a Model-Agnostic Meta-Learning-Physics-Informed Neural Networks (MAML-PINN) architecture and adaptive finite element discretization technology to optimize and control process parameters. The specific process is as follows.

[0072] Step S1: Utilizing a Physical Information Neural Network (PINN) to improve the framework, the processing progress deviation is used as a constraint. Electromagnetic transport equations, heat conduction equations, and fluid dynamics equations are coupled to perform coupled simulations of multiple physical fields, including the electric field distribution between silicon wafers, the temperature gradient during lamination, and the slurry flow behavior. Adaptive finite element discretization technology is then employed to quickly output high-precision simulation results and the effects of different parameter adjustment schemes on the progress correction, providing fundamental data for subsequent process parameter optimization.

[0073] Specifically, the electromagnetic transport equation, the heat conduction equation, and the fluid dynamics equation are as follows:

[0074] Electromagnetic transport equations:

[0075] Heat conduction equation:

[0076] Fluid dynamics equations:

[0077] in, For electric field strength, For density, Where is the dielectric constant. For temperature, For time, Thermal conductivity, For specific heat capacity, The internal heat source generation rate, For fluid velocity vector, This refers to kinematic viscosity.

[0078] The adaptive finite element discretization technique discretizes the computational domain into finite elements, and approximates the physical quantity u in spatial coordinate x as follows:

[0079]

[0080] in, Here, n represents the physical field function, and n is the number of element nodes. For shape functions, Let i be the node value at node i.

[0081] Step S2: Based on the MAML-PINN architecture, through a small number of batch experiments (<50 groups), the meta-learning algorithm is used to perform in-depth analysis on the data output by the multiphysics coupling simulation module to explore the potential process parameter correlations and optimization directions in the data, thereby quickly mastering the new process rules.

[0082] The specific processing steps based on the MAML-PINN architecture are as follows.

[0083] Suppose there are multiple tasks Each task has a corresponding dataset. For the PINN model In each task Perform an inner loop update:

[0084]

[0085] in, These are the task-specific model parameters obtained after performing inner loop learning for the j-th task. These are global, shared initial model parameters. For the inner loop learning rate, For the task The loss function.

[0086] Then, calculate the meta-loss. :

[0087]

[0088] in, To compensate for simulation error loss, For meta-learning loss, For defect-related losses, , , These are the weighting coefficients.

[0089] Then update the parameters through the outer loop. , The learning rate is the outer loop rate.

[0090] Using a small amount of experimental data, the meta-learning optimization module learns the new process rules and transmits the optimized process rules to the process disturbance self-correction module.

[0091] Step S3: The process disturbance self-correction module performs real-time compensation based on the process rules learned by the MAML-PINN architecture and in combination with disturbance factors such as ambient temperature drift (±2°C) and equipment aging (±2% accuracy drift) that are monitored in real time.

[0092] Specifically, the process disturbance self-correction module, based on the process rules transmitted from the meta-learning optimization module and combined with the real-time monitored environmental temperature drift, and equipment aging Perform parameter compensation.

[0093] Let the original process parameter vector be... The compensated parameter vector By compensation function Determined, that is:

[0094]

[0095] When the system detects deviations in process parameters or product defects, the meta-learning optimization module and the process disturbance self-correction module collaboratively adjust the parameters based on the simulation results output by the multiphysics coupled simulation module to minimize the total loss.

[0096] ,

[0097] in, To compensate for simulation error loss, For meta-learning loss, For defect-related losses, , , These are weighting coefficients; this ensures the stable and efficient operation of the process.

[0098] Figure 2-4 The diagram shows a comparison of the effects of process parameter optimization, real-time environmental temperature drift compensation, and cross-production line knowledge transfer in this embodiment.

[0099] In this invention, the MAML-PINN architecture and adaptive finite element discretization optimization algorithm play a core role in the collaborative optimization system of photovoltaic module stacking process parameters based on a digital twin of the process. In stage S1, the improved PINN framework under the MAML-PINN architecture couples multiphysics equations and uses adaptive finite element discretization technology to discretize the computational domain into finite elements, solve the discrete equations, and quickly output high-precision simulation results for key parameters such as electric field strength, temperature, and flow rate, providing data for the meta-learning module and assisting the system in preliminary analysis of process parameters. In stage S2, simulation data is received based on this architecture. For multiple tasks and datasets, the PINN model parameters are first updated in an inner loop, and then the outer loop updates the parameters after calculating the meta-loss. Through batch experimental data of fewer than 50 sets, the meta-learning algorithm is used to deeply analyze the data, uncover the correlation and optimization direction of process parameters, master new process rules, and pass them to the process disturbance self-correction module. In stage S3, the process disturbance self-correction module receives the rules and, combined with disturbances such as environmental temperature drift and equipment aging, as well as production and defect data, compensates for deviations in real time. This system leverages knowledge transfer to improve cross-production line adaptability, and uses defect tracing to pinpoint causes and provide feedback to optimize simulation models. When the system detects a problem, the two modules collaboratively adjust parameters to minimize total losses. Through dynamic parameter optimization, defect localization, improved adaptability, and waste avoidance, the system balances exploration and optimization, ensuring efficient operation in complex scenarios. This approach has effectively improved production efficiency and resource utilization in the photovoltaic module stacking process, demonstrating significant innovation and practicality.

[0100] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A photovoltaic module stacking process parameter collaborative optimization system based on process digital twin, characterized in that, include: The multiphysics coupling simulation module is used to construct a high-fidelity digital twin. By coupling electromagnetic transport equations, heat conduction equations and fluid dynamics equations, and using adaptive finite element discretization technology, it achieves efficient and high-precision simulation of photovoltaic module stacking process and outputs simulation results. The meta-learning optimization module is connected to the multiphysics coupling simulation module. It is used to receive the simulation results and, based on the meta-learning algorithm, learn and master new process rules through a small batch of experimental data, and output the optimized process rules. The process disturbance self-correction module is connected to the meta-learning optimization module and the multi-physics field coupled simulation module, respectively. It is used to receive the optimized process rules, and combine them with real-time production data to compensate for at least one disturbance factor in environmental temperature drift and equipment aging in real time, and to locate the cause of defects through process knowledge transfer and defect source analysis. The multiphysics coupled simulation module, meta-learning optimization module, and process disturbance self-correction module form a closed-loop optimization through a data-model-control three-flow coupling mechanism to dynamically adjust process parameters.

2. The photovoltaic module stacking process parameter collaborative optimization system based on process digital twin as described in claim 1, characterized in that, The multiphysics coupling simulation module is further configured as follows: Collect real-time data during the actual production process, including electric field distribution, temperature gradient, slurry flow behavior, equipment operating parameters, and defect feedback data. The real-time data is compared with a preset threshold, and a correction process is triggered when the preset deviation is exceeded. The data triggered during the correction process is fused with the electromagnetic transport equation, heat conduction equation, and fluid dynamics equation for simulation, and the simulation results are output to the meta-learning optimization module and the process disturbance self-correction module.

3. The photovoltaic module stacking process parameter collaborative optimization system based on process digital twin as described in claim 1, characterized in that, The meta-learning optimization module is further configured as follows: Receive the core parameter data, including electric field strength, temperature gradient and slurry flow rate, output by the multiphysics coupling simulation module; Using meta-learning algorithms, based on fewer than 50 sets of batch experimental data, we analyzed key process parameters including silicon wafer spacing, welding temperature, and paste concentration to master new process rules. The optimized process rules are passed to the process disturbance self-correction module for precise compensation of environmental temperature drift and equipment aging.

4. The photovoltaic module stacking process parameter collaborative optimization system based on process digital twin as described in claim 1, characterized in that, The process disturbance self-correction module is further configured as follows: The system receives the process rules transmitted by the meta-learning optimization module and, in conjunction with real-time production data and defect feedback data, performs real-time compensation for environmental temperature drift and equipment aging. By utilizing process knowledge transfer technology, optimized process knowledge acquired from one production line can be applied to other production lines to improve the success rate of cross-production line parameter adaptation. By tracing the source of defects, the probability distribution of the causes of lamination bubbles and cold solder joint defects is located, and the analysis results are fed back to the multiphysics coupling simulation module to optimize the simulation model.

5. The photovoltaic module stacking process parameter collaborative optimization system based on process digital twin as described in claim 1, characterized in that, The system is configured as follows: When deviations in process parameters or defects in products are detected, the meta-learning optimization module and the process disturbance self-correction module adjust the process parameters in a coordinated manner based on the simulation results output by the multi-physics coupling simulation module.

6. The photovoltaic module stacking process parameter collaborative optimization system based on process digital twin as described in claim 1, characterized in that, The system employs a Model Independent Meta-Learning-Physical Information Neural Network (MAML-PINN) architecture and adaptive finite element discretization technology to optimize and control process parameters, specifically including: Step S1: Using the improved framework of the Physical Information Neural Network (PINN), the processing progress deviation is used as a constraint condition. The electromagnetic transmission equation, heat conduction equation and fluid dynamics equation are coupled to perform multi-physics coupling simulation, and the simulation results are output using adaptive finite element discretization technology. Step S2: Based on the MAML-PINN architecture, the simulation results are processed through a meta-learning algorithm to uncover the correlation and optimization direction of process parameters and to master new process rules; Step S3: The process disturbance self-correction module performs real-time compensation based on the learned process rules and the environmental temperature drift and equipment aging disturbance factors monitored in real time.

7. The photovoltaic module stacking process parameter collaborative optimization system based on process digital twin as described in claim 6, characterized in that, In step S1, the electromagnetic transport equation, the heat conduction equation, and the fluid dynamics equation are specifically as follows: Electromagnetic transport equations: Heat conduction equation: Fluid dynamics equations: in, For electric field strength, For density, Where is the dielectric constant. For temperature, For time, Thermal conductivity, For specific heat capacity, The internal heat source generation rate, For fluid velocity vector, Kinematic viscosity; The adaptive finite element discretization technique discretizes the computational domain into finite elements, and approximates the physical quantity u in spatial coordinate x as follows: in, Here, n represents the physical field function, and n is the number of element nodes. For shape functions, Let i be the node value at node i.

8. The photovoltaic module stacking process parameter collaborative optimization system based on process digital twin as described in claim 6, characterized in that, In step S2, the processing based on the MAML-PINN architecture includes: Suppose there are multiple tasks Each task has a corresponding dataset. For the PINN model In each task Perform an inner loop update: in, These are the task-specific model parameters obtained after performing inner loop learning for the j-th task. These are global, shared initial model parameters. For the inner loop learning rate, For the task The loss function; Then, calculate the meta-loss. : in, To compensate for simulation error loss, For meta-learning loss, For defect-related losses, , , These are the weighting coefficients; Then update the parameters through the outer loop. , The learning rate is the outer loop rate.

9. A photovoltaic module stacking process parameter collaborative optimization system based on process digital twin as described in claim 6, characterized in that, In step S3, the process disturbance self-correction module, based on the process rules transmitted from the meta-learning optimization module and combined with the real-time monitored environmental temperature drift, and equipment aging Perform parameter compensation; Let the original process parameter vector be... The compensated parameter vector By compensation function Determined, that is: When the system detects deviations in process parameters or product defects, the meta-learning optimization module and the process disturbance self-correction module collaboratively adjust the parameters based on the simulation results output by the multiphysics coupled simulation module to minimize the total loss. ， in, To compensate for simulation error loss, For meta-learning loss, For defect-related losses, , , These are weighting coefficients; this ensures the stable and efficient operation of the process.

Images