A method and system for monitoring and early warning of vanadium redox flow battery stack health based on multi-physics field coupling and cloud edge collaboration

Abstract

This invention discloses a method and system for health monitoring and early warning of vanadium redox flow battery stacks based on multiphysics coupling and cloud-edge collaboration. The method includes: constructing a multiphysics coupling benchmark model with porous electrodes, where the electric and flow fields use a consistent grid in the porous electrode region; real-time acquisition of macroscopic operating parameters of the stack; real-time estimation of the macroscopic parameters as internal boundary conditions of the model using a neural network; rapid acquisition of the internal physical field distribution of the stack by substituting the boundary conditions into the local model; and accurate calculation of the diaphragm health index, flow channel blockage index, and thermal safety index based on the field information to achieve graded early warning. The system adopts a cloud-edge collaborative architecture: the multiphysics model and neural network are deployed locally, responsible for real-time online monitoring and inference, and storing real-time / near-real-time data; the cloud performs offline optimization of the neural network and reduced-order model based on long-term historical big data uploaded from the local side, and distributes the updated model back to the local side. The local side acquires parameters and outputs decision support information through an existing Supervisory Control and Data Acquisition (SCADA) system, which is not within the scope of this patent. This invention achieves online "transparent" perception of the internal state of the fuel cell stack and continuous self-evolution of the algorithm through a cloud-edge collaborative four-party architecture of "local real-time processing and slow optimization in the cloud". It can provide early warning of electrode aging, greatly improving the service life and operational safety of the fuel cell stack.

Description

Technical Field

[0001] This invention belongs to the field of safety monitoring technology for flow battery energy storage, specifically relating to a health status monitoring and early warning method and system for all-vanadium redox flow battery stacks based on multi-physics coupling and cloud-edge collaboration. Background Technology

[0002] Vanadium redox flow batteries (VRBs) offer advantages such as power-capacity decoupling, high safety, and long cycle life, making them promising candidates for large-scale energy storage. The fuel cell stack is the core power unit of a flow battery system, and its health directly impacts the performance, efficiency, and safety of the entire energy storage system. Problems such as separator aging, electrolyte mixing between positive and negative electrodes, flow channel blockage, and uneven heat distribution, especially material aging and performance degradation occurring at the microscopic level within porous electrodes, are the root causes of decreased stack capacity and even increased safety risks.

[0003] Currently, the monitoring methods for flow battery stacks in engineering are still relatively rudimentary. They typically only collect macroscopic operating parameters such as total voltage, total current, inlet and outlet temperatures, and system pressure, which are insufficient to effectively reflect the microscopic evolution of the stack's internal state, especially the core reaction region of porous electrodes. Existing early warning methods mostly rely on alarms for exceeding the threshold of a single parameter, which only respond after the fault has developed to a certain extent, making it difficult to identify potential hazards early.

[0004] Existing offline simulation and evaluation methods based on equivalent circuits or lumped parameters cannot reveal the long-term coupling effects of multiphysics fields and material properties within the internal space of the fuel cell stack. Furthermore, even though online monitoring technologies for multiphysics fields have been explored in some fields with extremely high safety requirements, their models and algorithms cannot be directly transferred to liquid phase, porous electrode flow battery systems.

[0005] More importantly, current technologies lack a system architecture that can effectively coordinate real-time online monitoring with offline algorithm optimization based on big data. Online monitoring has extremely high real-time requirements, while the training and updating of complex models consumes a large amount of computing resources and long-term historical data. If all tasks are placed on field devices, real-time performance and computing power cannot be simultaneously achieved; if everything is placed in the cloud, network latency and bandwidth limitations will severely impair the real-time performance of monitoring. Therefore, there is an urgent need for a system architecture and method that can balance real-time performance with the ability for continuous algorithm evolution. Summary of the Invention

[0006] This invention aims to provide a health monitoring and early warning method and system that can significantly improve the lifespan of vanadium redox flow batteries. Its core concept lies in constructing a "cloud-edge collaborative four-way architecture," deploying real-time-critical online monitoring and inference tasks on the local side (edge ​​computing side), and deploying offline algorithm optimization tasks that rely on long-term big data and massive computing resources on the cloud. On the local side, a deeply coupled multiphysics model is deployed to achieve online inversion from macroscopic measurable parameters to the internal physical field distribution of the battery stack, and to perform accurate health assessments and ultra-early warnings based on the physical field distribution. The cloud utilizes long-term operational data uploaded from the local side to continuously optimize the neural network and reduced-order model, and then distributes the updated model back to the local side, forming a closed loop of "local real-time processing, slow cloud optimization, and continuous model evolution."

[0007] The innovation of this invention at the architectural level lies in the aforementioned cloud-edge collaborative four-party architecture, which specifically includes: First, separation of responsibilities: the local side focuses on real-time online monitoring (millisecond to second-level response), while the cloud focuses on offline slow optimization (minute to hour-level iteration).

[0008] Second, data stratification: real-time or near-real-time running data is stored locally, while long-term historical big data is aggregated and stored in the cloud.

[0009] Third, bidirectional evolution: monitoring data is uploaded from the local side to the cloud, and optimized model parameters are sent from the cloud to the local side, enabling the algorithm to continuously evolve itself.

[0010] Fourth, system decoupling: the boundary between the system of this invention and the existing monitoring and data acquisition system (SCADA) on site is clearly defined. This invention focuses on the deep perception of health status and the accurate generation of early warning information.

[0011] The innovation of this invention at the physical model level lies in: First, deep coupling of multiple physics fields: establish a closed-loop coupling relationship between "electrode material (electrical conductivity / thermal conductivity / aging) - electric field - thermal field - flow field - external circuit", and use a consistent finite element mesh in the porous electrode region when solving the electric field and flow field to ensure the accurate spatial correspondence between the Joule heat distribution and the convective heat transfer boundary.

[0012] Second, electrode solid modeling: moving from one-dimensional equivalent circuits and lumped parameter models to solid models, accurately depicting the multi-physics field distribution inside porous electrodes.

[0013] Third, neural networks solve boundary problems: using a self-attention mechanism, focusing on the coupling boundaries of multiple physical fields that are most sensitive to aging and faults.

[0014] Through the above solutions, the present invention can achieve "transparent" perception of the internal state of the battery stack, and can provide early warning at the initial stage of electrode aging, thereby guiding precise operation and maintenance and greatly extending the cycle life of the vanadium redox flow battery stack.

[0015] The core inventive point of this invention lies in the fact that steps S1 to S5 are not isolated operations, but rather constitute a closed-loop collaborative technical link. Simultaneously, the collaboration between the local side and the cloud constitutes a higher-level system closed loop. Specifically: Step S1 provides a baseline model for step S4, step S2 provides input features for step S3, step S3 provides online unmeasurable internal boundary conditions for step S4, step S4 provides a judgment basis based on physical fields for step S5, and the decision support signal of step S5 guides the operator to adjust the stack operating status through the existing SCADA system, thereby affecting the acquisition parameters of step S2.

[0016] In step S3, the training data for the neural network must come from the same multiphysics coupling model constructed in step S1 to ensure that the physical meaning of the training data is consistent with the physical meaning when the boundary conditions are applied in step S4, thus achieving seamless integration between offline training and online application.

[0017] In step S4, each health assessment index is extracted from the physical field spatial distribution, rather than based on a simple threshold judgment of macroscopic parameters. This design makes the accuracy of health assessment highly dependent on the accuracy of the internal boundary conditions estimated in step S3 and the accuracy of the physical field distribution obtained in step S4, thus forming a cascade constraint of accuracy between each link in the technology chain.

[0018] At the system architecture level: the local side and the cloud form a cloud-edge collaborative four-party architecture: the local side processes online monitoring tasks and stores real-time data, while the cloud side processes offline optimization tasks and stores long-term data; the two sides achieve co-evolution of models and data through bidirectional data exchange. The multiphysics model on the local side ensures the real-time performance of online monitoring, while the algorithm optimization on the cloud side ensures that the system can continuously adapt to the long-term aging characteristics of the fuel cell stack.

[0019] The aforementioned synergistic effect enables this invention to achieve a holistic technological leap compared to existing isolated methods that rely solely on internal resistance change rates or use neural networks for fault classification or simulation error correction. Without adding internal sensors to the fuel cell stack, it can achieve online inversion of the internal physical field distribution and ultra-early identification of electrode micro-aging using only macroscopically measurable parameters. This overall technological effect is unpredictable by the sum of the effects of individual steps.

[0020] This invention aims to provide a health monitoring and early warning method and system that can significantly improve the lifespan of vanadium redox flow batteries. Its core concept lies in constructing a "cloud-edge collaborative four-way architecture," deploying real-time-critical online monitoring and inference tasks on the local side (edge ​​computing side), and deploying offline algorithm optimization tasks that rely on long-term big data and massive computing resources on the cloud. On the local side, a deeply coupled multiphysics model is deployed to achieve online inversion from macroscopic measurable parameters to the internal physical field distribution of the battery stack, and to perform accurate health assessments and ultra-early warnings based on the physical field distribution. The cloud utilizes long-term operational data uploaded from the local side to continuously optimize the neural network and reduced-order model, and then distributes the updated model back to the local side, forming a closed loop of "local real-time processing, slow cloud optimization, and continuous model evolution."

[0021] The innovation of this invention at the architectural level lies in the aforementioned cloud-edge collaborative four-party architecture, which specifically includes: First, separation of responsibilities: the local side focuses on real-time online monitoring (millisecond to second-level response), while the cloud focuses on offline slow optimization (minute to hour-level iteration).

[0022] Second, data stratification: real-time or near-real-time running data is stored locally, while long-term historical big data is aggregated and stored in the cloud.

[0023] Third, bidirectional evolution: monitoring data is uploaded from the local side to the cloud, and optimized model parameters are sent from the cloud to the local side, enabling the algorithm to continuously evolve itself.

[0024] Fourth, system decoupling: the boundary between the system of this invention and the existing monitoring and data acquisition system (SCADA) on site is clearly defined. This invention focuses on the deep perception of health status and the accurate generation of early warning information.

[0025] The innovation of this invention at the physical model level lies in: First, deep coupling of multiple physics fields: establish a closed-loop coupling relationship between "electrode material (electrical conductivity / thermal conductivity / aging) - electric field - thermal field - flow field - external circuit", and use a consistent finite element mesh in the porous electrode region when solving the electric field and flow field to ensure the accurate spatial correspondence between the Joule heat distribution and the convective heat transfer boundary.

[0026] Second, electrode solid modeling: moving from one-dimensional equivalent circuits and lumped parameter models to solid models, accurately depicting the multi-physics field distribution inside porous electrodes.

[0027] Third, neural networks solve boundary problems: using a self-attention mechanism, focusing on the coupling boundaries of multiple physical fields that are most sensitive to aging and faults.

[0028] Through the above solutions, the present invention can achieve "transparent" perception of the internal state of the battery stack, and can provide early warning at the initial stage of electrode aging, thereby guiding precise operation and maintenance and greatly extending the cycle life of the vanadium redox flow battery stack.

[0029] The core inventive point of this invention lies in the fact that steps S1 to S5 are not isolated operations, but rather constitute a closed-loop collaborative technical link. Simultaneously, the collaboration between the local side and the cloud constitutes a higher-level system closed loop. Specifically: Step S1 provides a baseline model for step S4, step S2 provides input features for step S3, step S3 provides online unmeasurable internal boundary conditions for step S4, step S4 provides a judgment basis based on physical fields for step S5, and the decision support signal of step S5 guides the operator to adjust the stack operating status through the existing SCADA system, thereby affecting the acquisition parameters of step S2.

[0030] In step S3, the training data for the neural network must come from the same multiphysics coupling model constructed in step S1 to ensure that the physical meaning of the training data is consistent with the physical meaning when the boundary conditions are applied in step S4, thus achieving seamless integration between offline training and online application.

[0031] In step S4, each health assessment index is extracted from the physical field spatial distribution, rather than based on a simple threshold judgment of macroscopic parameters. This design makes the accuracy of health assessment highly dependent on the accuracy of the internal boundary conditions estimated in step S3 and the accuracy of the physical field distribution obtained in step S4, thus forming a cascade constraint of accuracy between each link in the technology chain.

[0032] At the system architecture level: the local side and the cloud form a cloud-edge collaborative four-party architecture: the local side processes online monitoring tasks and stores real-time data, while the cloud side processes offline optimization tasks and stores long-term data; the two sides achieve co-evolution of models and data through bidirectional data exchange. The multiphysics model on the local side ensures the real-time performance of online monitoring, while the algorithm optimization on the cloud side ensures that the system can continuously adapt to the long-term aging characteristics of the fuel cell stack. The aforementioned synergistic effect enables this invention to achieve a holistic technological leap compared to existing isolated methods that rely solely on internal resistance change rates or use neural networks for fault classification or simulation error correction. Without adding internal sensors to the fuel cell stack, it can achieve online inversion of the internal physical field distribution and ultra-early identification of electrode micro-aging using only macroscopically measurable parameters. This overall technological effect is unpredictable by the sum of the effects of individual steps. Attached Figure Description

[0033] This application document includes the following two accompanying drawings: Figure 1 : This is a schematic diagram of the overall architecture of the method and system of the present invention. Figure 1 The diagram uses a layered architecture to clearly illustrate the two levels of "cloud" and "local side," and their relationship with the "on-site physical system" and the "existing SCADA system." The cloud includes a large-capacity database (storing long-term historical data) and an algorithm optimization module (responsible for offline model training and updates). The local side (edge ​​computing devices) includes multiphysics coupling models / reduced-order model response surfaces, neural network models, fusion evaluation and hierarchical early warning modules, and stores real-time / near-real-time data. Arrows clearly indicate bidirectional data exchange: the local side uploads real-time / near-real-time monitoring data to the cloud, and the cloud distributes the optimized model to the local side. The local side obtains macroscopic operating parameters from the existing SCADA system on-site via industrial communication protocols and outputs the generated early warning signals and decision support information to that SCADA system. Dashed boxes and annotations clearly indicate that the SCADA system and its data interface with the local side are not within the protection scope of this system. Figure 2 : This is a block diagram of the coupling of multiple physics fields of electricity, heat, flow and aging. Figure 2 Using the central porous electrode material as the core object, its key attribute parameters (electrical conductivity σ, thermal conductivity k, porosity ε) are labeled. Four physical field modules are drawn around the core object: an electric field model, a flow field model, a thermal field model, and a material aging model. Each module is connected to the porous electrode material via arrows to illustrate the following coupling relationships: the porous electrode material provides the electric field model with the electrical conductivity σ distribution, the flow field model with the permeability distribution, and the thermal field model with the thermal conductivity k distribution; the electric field model transfers the volumetric heat source to the thermal field model through Joule heat q; the flow field model transfers the heat dissipation boundary conditions to the thermal field model through the convective heat transfer coefficient h; the temperature distribution T output by the thermal field model is fed back to the material aging model, driving the calculation of the aging function; the material aging model feeds back the updated material properties, such as electrical conductivity σ(t) and thermal conductivity k(t), to the porous electrode material, forming a closed-loop feedback at the material property level. Figure 2 The paper also specifies the consistency requirements for the finite element mesh between the electric field model and the flow field model.

Claims

1. A method for health monitoring and early warning of vanadium redox flow battery stacks based on multiphysics coupling and cloud-edge collaboration, characterized in that, Includes the following steps: Step S1: Construct a multiphysics coupled benchmark model: Establish an electro-thermal-fluid-aging multiphysics coupled model of the target vanadium redox flow battery stack; wherein, the model performs geometric modeling and mesh generation of the porous electrode regions of the positive and negative electrodes in the stack, and runs in a general multiphysics simulation software under standard rated conditions to obtain the benchmark parameter field under healthy conditions; the standard rated conditions are defined as: the stack operates at rated current Operation, electrolyte at rated flow rate Supply and ambient temperature are 25℃, and the state of charge (SOC) of the fuel cell stack is 50%. Step S2: Obtain online measured parameters: During the operation of the fuel cell stack, the macroscopic operating parameters of the fuel cell stack are collected in real time. The macroscopic operating parameters include at least the total voltage. Total current Inlet and outlet pressure difference Total flow Average temperature of fuel cell stack and the rate of change of outlet temperature ; Step S3: Boundary parameter estimation based on neural network: The collected macroscopic operating parameters are input into a pre-trained multi-layer neural network; the multi-layer neural network uses the macroscopic operating parameters as the input layer and the internal boundary condition parameters of each physical field in the multi-physics coupling model as the output layer, and adopts a self-attention mechanism to improve the ability to extract features at the physical field coupling boundary related to specific aging or failure modes. Step S4: Model-driven multi-index calculation: The internal boundary condition parameters estimated by the neural network are loaded into the multi-physics coupled benchmark model deployed on the local side. The physical field distribution inside the stack is obtained by quickly solving or querying the pre-established reduced-order model response surface online. Based on the physical field distribution, multiple health assessment indices characterizing diaphragm health, flow channel blockage, thermal safety, and overall performance are calculated. Step S5: Tiered Early Warning and Decision Support: Based on the calculated health assessment indices, tiered early warning is implemented, and decision support signals are output. The key feature is that steps S1 to S5 constitute a closed-loop collaborative technical link: step S1 provides a benchmark model for loading internal boundary condition parameters for step S4; step S2 provides input feature parameters for step S3; step S3 provides online, non-measurable internal boundary condition parameters for step S4; step S4 loads the internal boundary condition parameters into the benchmark model to obtain the internal physical field distribution of the fuel cell stack, and calculates a health assessment index based on the physical field distribution, providing a basis for graded early warning for step S5; step S5 outputs a decision support signal, which guides the adjustment of the fuel cell stack's operating status through the existing monitoring and data acquisition system, thereby affecting the macroscopic operating parameters collected in step S2, forming an online closed-loop collaboration from monitoring to decision support, and then from operation feedback to monitoring.

2. The method according to claim 1, characterized in that, The coupling relationship of the electro-thermal-fluid-aging multiphysics coupling model in step S1 specifically includes: Electrode material-electric field-circuit coupling: The conductivity distribution of the porous electrode material is calculated using finite element method to obtain the potential distribution in the electrode domain, and then coupled with the equivalent circuit model of the fuel cell stack to calculate the total voltage and internal current density distribution of the fuel cell stack. Electrode material-thermal field-flow field coupling: The thermal conductivity distribution of the porous electrode material is coupled with the reaction heat source, and with the lumped parameter thermal model and fluid model to calculate the temperature distribution and flow distribution inside the stack; Electrode material aging-thermal field-electric field coupling: An aging model for electrode materials is established, which defines the decay function of material conductivity and thermal conductivity with the number of cycles or time, and feeds back the material properties after aging to the solution of electric field and thermal field in real time; The electric field model and the flow field model use the same finite element mesh in the porous electrode region to ensure the accurate spatial correspondence between the Joule heat distribution and the convective heat transfer boundary.

3. The method according to claim 1, characterized in that, The multilayer neural network in step S3 employs a self-attention mechanism, with its attention weights focused on a preset high-sensitivity coupling boundary in the multiphysics coupling model. The preset high-sensitivity coupling boundary includes at least: an ionic conductivity boundary layer at the interface between the diaphragm and the electrode; a contact resistance boundary layer at the interface between the electrode and the bipolar plate; and a convective heat transfer coefficient boundary layer at the flow channel wall.

4. The method according to claim 1, characterized in that, The health assessment index in step S4 includes: Diaphragm health index Based on the potential distribution obtained from the online solution, the voltage standard deviation of each individual cell is calculated. And in conjunction with internal resistance, calculate ,in and As a health benchmark, and These are the weighting coefficients; Flow channel blockage index Based on the flow field distribution obtained from online solutions, low-velocity regions with flow velocities below a certain percentage of the preset healthy baseline velocity are identified, and this is combined with the inlet and outlet pressure difference. With traffic ,calculate ,in and As a health benchmark; Thermal safety index Based on the temperature distribution obtained from online solutions, local hotspots with temperatures exceeding a certain threshold of a preset healthy baseline temperature and their temperature rise rates are identified. Combined with the overall temperature rise, calculations are performed. ,in This is the highest temperature inside the fuel cell stack. The preset maximum safe operating temperature, This is the factor affecting the rate of temperature rise.

5. The method according to claim 2, characterized in that, The aging model of the electrode material is used to simulate the surface oxidation and etching of the carbon felt electrode of the vanadium redox flow battery under long-term acidic electrolyte environment. This aging process leads to an increase in electrode contact resistance. Increase and active specific surface area The aging function is expressed as: ,in For the equivalent number of iterations, The initial contact resistance. and This represents the aging factor.

6. The method according to claim 1, characterized in that, In step S4, the physical field distribution is quickly obtained by querying the pre-established reduced-order model response surface. The reduced-order model response surface is constructed by interpolating or fitting a large number of offline simulation results of the multi-physics coupling model under different boundary conditions.

7. A health monitoring and early warning system for an all-vanadium redox flow battery stack based on multiphysics coupling and cloud-edge collaboration, characterized in that, Adopting a local-cloud collaborative architecture, including the local side and the cloud: The local side is deployed at the fuel cell stack operation site or in an adjacent edge computing device. It is used to process online monitoring tasks with real-time requirements exceeding a preset threshold and to store real-time or near-real-time operational data. The local side includes: The model building module is used to build the multiphysics coupling benchmark model as described in claim 2. This model is mainly deployed and run on the local side. The data acquisition module is used to obtain macroscopic operating parameters in real time from the existing monitoring and data acquisition system at the fuel cell stack site via industrial communication protocols; The neural network estimation module is used to implement the online estimation of boundary parameters as described in claim 3; The online solver / query module is used to drive the local model to calculate the physical field distribution; The health index calculation module is used to calculate various health assessment indices; The early warning decision support module is used to generate graded early warning signals and decision support information based on the calculated health assessment indices, and output the decision support information to the existing monitoring and data acquisition system. The cloud platform is deployed on a remote server or cloud computing platform to process offline optimization tasks that consume computing resources higher than a preset threshold or require long-term historical data, and to store historical data of the fuel cell stack's long-term operation. The cloud platform includes: A large-capacity storage module is used to store historical macroscopic parameters, historical internal boundary condition parameters, and historical health indices of the fuel cell stack that are uploaded from the local side during long-term operation. The algorithm optimization module is used to perform offline training or update optimization of the response surface of the neural network and / or the reduced-order model in the local side using long-term historical data in the large-capacity storage module. There is bidirectional data exchange between the local side and the cloud: the local side uploads real-time or near real-time operating data and health assessment results to the cloud for storage; the cloud sends optimized and updated neural network model parameters and / or reduced-order model response surfaces to the local side. Furthermore, the existing monitoring and data acquisition systems and their integration methods with this system are not within the scope of protection of this system.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 6.

9. The method according to claim 1, characterized in that, The specific coordination relationship between step S3 and step S4 is as follows: The training data for the multilayer neural network in step S3 is generated in the following way: on the multiphysics coupling benchmark model constructed in step S1, the values ​​of each internal boundary condition parameter are systematically changed, the model is run and its corresponding macroscopic operating parameters are recorded to form a "macroscopic operating parameter - internal boundary condition parameter" mapping dataset. The multi-layer neural network is trained using this mapping dataset to learn the macro-micro mapping relationship; The data structure of the training data is consistent with the data structure of the macroscopic operating parameters obtained in step S2, and the physical meaning of the training data is consistent with the physical meaning of the internal boundary condition parameters loaded in step S4, so that the training data generated by steps S3 and S4 based on the same physical model can be seamlessly connected.

10. The method according to claim 1, characterized in that, There is a deep, physically driven correlation between the health assessment index in step S4 and the physical field distribution: The diaphragm health index Calculate the voltage standard deviation of each individual cell based on the potential distribution in the aforementioned physical field distribution. and combined with internal resistance Weighting is applied to reflect the impact of diaphragm aging on the electric field distribution inside the fuel cell stack; The flow channel blockage index Low-velocity regions are identified based on the flow field distribution within the physical field distribution, and this is combined with the inlet and outlet pressure difference. With traffic Calculations were performed to reflect the impact of flow channel blockage on the internal flow field distribution of the fuel cell stack; The thermal safety index Local hotspots and their temperature rise rates are identified based on the temperature distribution within the physical field distribution. To reflect the impact of thermal anomalies on the temperature field distribution inside the fuel cell stack; Each health assessment index is not based on a simple threshold judgment of macroscopic operating parameters, but rather on spatially interpretable distribution features extracted from the physical field distribution. This makes the accuracy of the health assessment dependent on the accuracy of the internal boundary conditions estimated in step S3 and the accuracy of the physical field distribution obtained in step S4, thus forming a constraint on the dependence of index accuracy on the preceding links in the technology chain.

11. The method according to claim 1, characterized in that, The reduced-order model response surface pre-established in step S4 is constructed by multidimensional interpolation or data fitting based on a large number of offline simulation results of the multiphysics coupling model under different internal boundary conditions. This ensures that the response time of online queries meets the real-time requirements of the stack operation, thereby guaranteeing the online real-time performance of the closed-loop collaborative technology link in engineering applications.

12. The system according to claim 7, characterized in that, The specific collaborative relationship between the local side and the cloud is as follows: The local side is responsible for handling online monitoring tasks with real-time requirements exceeding a preset threshold, including real-time acquisition of macroscopic operating parameters, online estimation of internal boundary conditions, online solution or query of physical field distribution, calculation of health assessment index and generation of graded early warning signals, and storage of real-time or near-real-time operating data. The cloud is responsible for processing offline optimization tasks that consume computing resources higher than a preset threshold or that require reliance on long-term historical data, including training neural network models based on long-term big data, reconstructing and updating the response surface of reduced-order models, and storing the long-term historical data. The data exchange between the local side and the cloud adopts an asynchronous communication mechanism, and the online monitoring function of the local side is not interrupted due to the data exchange delay with the cloud.

Images