Hydrogen fuel cell production management method and system based on digital twin technology

Abstract

The invention discloses a hydrogen fuel cell production management method and system based on a digital twinning technology, and relates to the technical field of digital twinning, and the method comprises the steps: collecting the production data of a hydrogen fuel cell in real time, carrying out the data preprocessing, and generating standardized data; based on the standardized data, constructing a digital twinborn model by utilizing finite element analysis and a computational fluid dynamics strategy, and training the digital twinborn model through historical data; building a reinforcement learning environment based on the trained digital twinborn model, defining a reinforcement learning agent, and generating an optimal control strategy; production parameters are adjusted according to the optimal control strategy, production data are fed back through the digital twinborn model, the optimization effect is verified regularly, and the digital twinborn model and the optimal control strategy are iteratively updated. According to the method, the optimal strategy is autonomously learned in the simulation environment through the reinforcement learning agent, the generated control strategy can dynamically adjust production parameters, efficiency is improved, energy consumption is reduced, quality is improved, and intelligent decision making and control are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of digital twins, and particularly to a hydrogen fuel cell production management method and system based on digital twin technology. Background Art

[0002] As an efficient and clean energy conversion technology, hydrogen fuel cells have received extensive attention and rapid development in the new energy field in recent years; with the rise of the hydrogen energy industry, the production management of hydrogen fuel cells has become a key link to improve production efficiency, reduce costs, and improve product quality; traditional production management methods mainly rely on automated control systems and manual experience, although to a certain extent, they can ensure the stable operation of the production process, but in the face of complex production environments and dynamic changes, it is often difficult to achieve precise optimization control and intelligent decision-making; Existing hydrogen fuel cell production management systems have deficiencies in dynamic optimization control and intelligent decision-making support, and it is difficult to achieve the efficient operation and precise optimization of the production process; for example, traditional methods cannot adjust production parameters in real time to adapt to dynamic changes in the production process, resulting in problems such as low production efficiency, high energy consumption, and unstable product quality; in addition, existing technologies lack the ability to comprehensively monitor and predict the production process, and cannot detect potential faults and optimization spaces in a timely manner. Summary of the Invention

[0003] In view of the above existing problems, the present invention is proposed.

[0004] Therefore, the present invention provides a hydrogen fuel cell production management method based on digital twin technology to solve the deficiencies in dynamic optimization control and intelligent decision-making support.

[0005] To solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides a hydrogen fuel cell production management method based on digital twin technology, which includes: collecting production data of hydrogen fuel cells in real time, performing data preprocessing to generate standardized data; based on the standardized data, constructing a digital twin model using finite element analysis and computational fluid dynamics strategies, and training the digital twin model with historical data; constructing a reinforcement learning environment based on the trained digital twin model, defining a reinforcement learning agent, and generating an optimal control strategy; adjusting production parameters according to the optimal control strategy, feeding back production data through the digital twin model, regularly verifying the optimization effect, and iteratively updating the digital twin model and the optimal control strategy.

[0006] As a preferred embodiment of the hydrogen fuel cell production management method based on digital twin technology of the present invention, wherein: the production data includes temperature, pressure, gas flow rate, current density, equipment status, quality inspection results, and operation logs; The preprocessing is to clean, standardize, extract features and reduce the dimension of the data to generate standardized data.

[0007] As a preferred solution of the hydrogen fuel cell production management method based on digital twin technology described in the present invention, wherein: based on the standardized data, a digital twin model is constructed by using finite element analysis and computational fluid dynamics strategies. The specific steps are as follows. According to the production data, the input, output and internal structure of the digital twin model are defined to generate the architecture of the initial digital twin model. Based on the standardized data and the architecture of the initial digital twin model, a multi-physics field model is constructed according to the structural mechanics problems and fluid dynamics problems in the production process. By using the mesh generation method, the multi-physics field model is divided to generate finite element meshes and computational fluid dynamics meshes. By using a coupling algorithm, the finite element analysis meshes and computational fluid dynamics meshes are combined to generate a digital twin model.

[0008] As a preferred solution of the hydrogen fuel cell production management method based on digital twin technology described in the present invention, wherein: the digital twin model is trained by using historical data. The specific steps are as follows. The historical data is divided into a training set, a validation set and a test set. The digital twin model is initialized by using the Xavier initialization method. The digital twin model is trained by using the training set, the hyperparameters of the digital twin model are adjusted by using the validation set, and the performance of the digital twin model is evaluated by using the test set to generate a trained digital twin model. The parameters and structure of the digital twin model are dynamically adjusted by using a machine learning model to generate an updated digital twin model.

[0009] As a preferred solution of the hydrogen fuel cell production management method based on digital twin technology described in the present invention, wherein: a reinforcement learning environment is constructed based on the trained digital twin model. The specific steps are as follows. Based on the updated digital twin model, the operation data and dynamic characteristics of the production process are analyzed to define the state space and the action space. Based on the historical production data and performance indicators, an optimization target is generated through the analysis of the digital twin model, and a reward function is defined in combination with the dynamic characteristics. Based on the state space, the action space and the reward function, a reinforcement learning environment is constructed by using simulation environment integration and interface development.

[0010] As a preferred solution of the hydrogen fuel cell production management method based on digital twin technology described in the present invention, wherein: the reinforcement learning agent is defined to generate an optimal control strategy. The specific steps are as follows. Based on the reinforcement learning algorithm, a neural network architecture is defined using the basic fully connected network method to construct a policy network and a value network; Based on the policy network and the value network, a policy update mechanism is defined through the Actor-Critic method; The policy network, the value network and the policy update mechanism are combined to generate a reinforcement learning agent; Through interaction training between the reinforcement learning environment and the reinforcement learning agent, interaction data is generated, and the policy network is optimized through the policy update mechanism to generate an optimal control strategy.

[0011] As a preferred solution of the hydrogen fuel cell production management method based on digital twin technology according to the present invention, wherein: adjusting the production parameters according to the optimal control strategy, feeding back the production data through the digital twin model, regularly verifying the optimization effect, and iteratively updating the digital twin model and the optimal control strategy. The specific steps are as follows. Based on the optimal control strategy, dynamically adjust the equipment parameters in the hydrogen fuel cell production process, and combine the digital twin model to real-time feedback the production data; Regularly compare the actual production data with the optimization target, evaluate the optimization effect of the optimal control strategy on production efficiency, energy consumption and product quality, and generate a verification result; Based on the verification result, update the digital twin model and the optimal control strategy.

[0012] In a second aspect, the present invention provides a hydrogen fuel cell production management system based on digital twin technology, including a data acquisition module, a model construction module, a policy generation module and an optimization module; the data acquisition module is used to collect the production data of the hydrogen fuel cell in real time, perform data preprocessing, and generate standardized data; the model construction module is used to construct a digital twin model based on the standardized data by using the finite element analysis and computational fluid dynamics strategy, and train the digital twin model through historical data; the policy generation module is used to construct a reinforcement learning environment based on the trained digital twin model, define a reinforcement learning agent, and generate an optimal control strategy; the optimization module is used to adjust the production parameters according to the optimal control strategy, feed back the production data through the digital twin model, regularly verify the optimization effect, and iteratively update the digital twin model and the optimal control strategy.

[0013] In a third aspect, the present invention provides a computer device, including a memory and a processor, wherein: when the computer program is executed by the processor, any step of the hydrogen fuel cell production management method based on digital twin technology as described in the first aspect of the present invention is implemented.

[0014] Fourthly, the present invention provides a computer-readable storage medium, on which a computer program is stored, wherein: when the computer program is executed by a processor, any step of the hydrogen fuel cell production management method based on digital twin technology as described in the first aspect of the present invention is implemented.

[0015] The beneficial effects of the present invention are as follows: The digital twin model constructed by finite element analysis and computational fluid dynamics strategy can accurately capture the mechanical and fluid characteristics in the production of hydrogen fuel cells, reflect the actual production situation, simulate physical phenomena, provide a basis for the construction of the reinforcement learning environment and the generation of optimal control strategies, and help optimize production; The reinforcement learning agent autonomously learns the optimal strategy in the simulated environment, and the generated control strategy can dynamically adjust production parameters, improve efficiency, reduce energy consumption, improve quality, and achieve intelligent decision-making and control. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0017] Figure 1 It is a flowchart of the hydrogen fuel cell production management method based on digital twin technology in the present invention.

[0018] Figure 2 It is a schematic diagram of the hydrogen fuel cell production management system based on digital twin technology in the present invention.

[0019] Figure 3 It is a flowchart of the construction of the digital twin model of the hydrogen fuel cell production management method based on digital twin technology in the present invention.

[0020] Figure 4 It is a flowchart of the construction of the reinforcement learning environment of the hydrogen fuel cell production management method based on digital twin technology in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] In order to make the above objects, features, and advantages of the present invention more obvious and understandable, the following will make a detailed description of the specific embodiments of the present invention with reference to the accompanying drawings of the specification.

[0022] In the following description, many specific details are set forth to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0023] Secondly, the so-called "one embodiment" or "embodiment" herein refers to specific features, structures, or characteristics that may be included in at least one implementation of the present invention. The appearances of "in one embodiment" in different places in this specification do not all refer to the same embodiment, nor are they separate or alternative embodiments that are mutually exclusive of other embodiments.

[0024] Embodiment 1, referring to Figures 1 to 4 , is the first embodiment of the present invention. This embodiment provides a production management method for hydrogen fuel cells based on digital twin technology, including the following steps: S1. Real-time collect the production data of hydrogen fuel cells, perform data preprocessing, and generate standardized data; The production data includes temperature, pressure, gas flow rate, current density, equipment status, quality inspection results, and operation logs; It should be noted that the production data of hydrogen fuel cells covers temperature, pressure, gas flow rate, current density, equipment status, quality inspection results, and operation logs. The production data of hydrogen fuel cells is obtained through real-time collection by sensors, monitoring by automated detection equipment, and manual recording. After data acquisition and monitoring preprocessing, standardized data is generated. They comprehensively reflect the physical quantity changes, equipment operation status, product quality characteristics, and manual operation information in the production process, providing a basis for real-time monitoring, analysis, and optimization of the production process.

[0025] The preprocessing is to clean, standardize, extract features, and reduce the dimension of the data to generate standardized data.

[0026] Furthermore, the preprocessing of the production data of hydrogen fuel cells includes data cleaning, standardization, feature extraction, and dimension reduction. Data cleaning removes noise and outliers and fills in missing values; data standardization makes the data have the same dimension through normalization or standardization processing; feature extraction extracts features that affect the performance indicators of the production process (such as production efficiency, energy consumption, product quality, etc.) from high-dimensional data to reduce the dimension and improve the model efficiency; feature dimension reduction further reduces the data dimension using techniques such as PCA or LDA, retaining the main information.

[0027] S2. Based on the standardized data, construct a digital twin model using finite element analysis and computational fluid dynamics strategies, and train the digital twin model with historical data; Define the input, output, and internal structure of the digital twin model according to the production data, and generate the architecture of the initial digital twin model; It should be noted that the internal structure is a core component of the digital twin model, which defines how the digital twin model processes input data and generates output results. The internal structure includes the input layer, hidden layer, output layer, internal connections, and data flow, as well as the parameters and hyperparameters of the digital twin model. By reasonably defining the internal structure, it can be ensured that the digital twin model can accurately reflect the physical phenomena and dynamic characteristics in the production process, providing strong support for the optimization and control of the production process.

[0028] The specific process is as follows. First, define the input layer of the digital twin model according to the extracted features that affect the performance indicators of the production process to ensure data compatibility and the stability of the digital twin model; then define the output layer according to the target output of the digital twin model to make it correspond to the actual production indicators; then define the internal structure of the digital twin model, including the number of hidden layers, the number of neurons, and the selection of activation functions; finally, integrate to generate the architecture of the initial digital twin model and perform parameter initialization to ensure the stability of the digital twin model training.

[0029] Based on the standardized data and the architecture of the initial digital twin model, construct a multi-physics field model according to the structural mechanics problems and fluid dynamics problems in the production process; Furthermore, first through data preparation and analysis, after data preprocessing combined with preliminary analysis, correlation analysis, time series analysis, and fault analysis, it is possible to identify the stress, strain, and deformation problems that may exist during the operation of the equipment, as well as the flow characteristics, pressure distribution, and flow rate changes of the fluid in the equipment; then construct a structural mechanics model and a fluid dynamics model respectively, including defining the physical model, mesh generation, and physical model solution; then integrate the structural mechanics model and the fluid dynamics model through data interaction and coupling algorithms to generate a complete multi-physics field model; finally, ensure the accuracy and reliability of the multi-physics field model through verification and optimization. These steps together constitute the construction process from standardized data to a complete multi-physics field model, providing a solid foundation for subsequent production management optimization.

[0030] Through the mesh generation method, divide the multi-physics field model to generate finite element meshes and fluid dynamics meshes; The specific process is as follows. First, import and preprocess the multi-physics field model to ensure the integrity of the multi-physics field model and the identification of features; then select a suitable mesh generation strategy according to the geometric complexity and physical field characteristics of the multi-physics field model; then perform mesh generation on the structural mechanics part and the fluid dynamics part respectively, and conduct quality inspection and optimization; finally, export the optimized mesh data and integrate it into the multi-physics field model. This series of steps ensures that the generated meshes can accurately reflect the geometric complexity and physical field characteristics of the multi-physics field model, providing a high-quality foundation for subsequent numerical simulation and analysis.

[0031] Through a coupling algorithm, the finite element analysis mesh and the computational fluid dynamics mesh are combined to generate a digital twin model.

[0032] Furthermore, first, data extraction and format unification are performed on the finite element analysis mesh and the computational fluid dynamics mesh; then, interpolation methods (such as conservative interpolation and non-conservative interpolation) are used to map the loads in the computational fluid dynamics mesh to the mesh of the finite element analysis mesh, and at the same time, the displacements or velocities of the finite element analysis mesh are fed back into the computational fluid dynamics mesh. For example, the nearest neighbor interpolation or linear interpolation method can be used to map the pressure distribution in the computational fluid dynamics mesh to the mesh nodes of the finite element analysis mesh to define the coupling interface, ensuring the accurate transfer of data between the finite element analysis mesh and the computational fluid dynamics mesh; then, a strong coupling algorithm is used and parameter settings are made; subsequently, the finite element analysis mesh and the computational fluid dynamics mesh are integrated through the coupling algorithm, and the structural mechanics equations, fluid dynamics equations, and coupling equations are solved; finally, the accuracy and reliability of the generated digital twin model are ensured through result verification and optimization. This series of steps together constitute the construction process from two independent meshes to a complete digital twin model, providing a solid foundation for subsequent production management optimization.

[0033] The historical data is divided into a training set, a validation set, and a test set; It should be noted that historical data refers to various data accumulated during past production processes, including production data, performance index data, environmental data, maintenance and repair records, production plans and scheduling data, and other relevant data. These data reflect various states, operations, and results in the production process and are an important basis for constructing and training digital twin models. By reasonably dividing historical data into a training set, a validation set, and a test set, the performance of the digital twin model can be effectively trained and evaluated, ensuring the accuracy and reliability of the digital twin model in practical applications.

[0034] The specific process is as follows: first, the historical data is cleaned and standardized to ensure data integrity and consistency; then, according to a preset ratio (such as 70%, 15%, 15%), the data is divided into a training set, a validation set, and a test set using a random sampling method, and the subset index of each data point is recorded; finally, through steps such as distribution inspection, sample representativeness verification, and data volume inspection, it is ensured that each subset can accurately represent the characteristics and distribution of the overall data. This series of steps provides a scientific and reasonable data foundation for subsequent digital twin model training, parameter adjustment, and performance evaluation, ensuring the reliability and effectiveness of the model.

[0035] Initialize the digital twin model using the Xavier initialization method, train the digital twin model using the training set, adjust the hyperparameters of the digital twin model using the validation set, evaluate the performance of the digital twin model using the test set, and generate the trained digital twin model; It should be noted that the Xavier initialization method, also known as Glorot initialization, is a technique for initializing the weights of neural networks. It was proposed by Xavier Glorot and Yoshua Bengio in 2010 to address the problem of vanishing or exploding gradients during the training of deep feedforward neural networks. The core idea of this method is to reasonably set the initial values of the weights according to the number of inputs (fan_in) and outputs (fan_out) of each layer of neurons, so that the variance of the activation function is approximately the same during forward and backward propagation, thus maintaining the effective flow of signals.

[0036] The specific process is as follows. Initializing the digital twin model using the Xavier initialization method can provide a good start for the training of the digital twin model and avoid training difficulties caused by unreasonable initial parameters. During the training process, make full use of the training set data, and through reasonable definition and optimization strategies of the digital twin model structure, enable the digital twin model to learn effective feature representations. Adjust the hyperparameters using the validation set to ensure that the digital twin model has good adaptability and generalization ability in complex and changing actual environments. Finally, through the strict evaluation of the test set, verify the performance and reliability of the digital twin model, and generate a fully trained and optimized digital twin model to provide strong support for practical applications.

[0037] Dynamically adjust the parameters and structure of the digital twin model through a machine learning model to generate an updated digital twin model.

[0038] Furthermore, first, train a machine learning model using the preprocessed data. Select a suitable machine learning algorithm according to the specific task and data characteristics, such as neural networks, support vector machines, etc., and adjust the parameters of the machine learning model through an optimization algorithm so that the machine learning model can accurately learn the patterns and features in the data. Then, use the trained machine learning model to dynamically adjust the parameters of the digital twin model, predict the optimal values of each parameter in the digital twin model and update them to improve the accuracy and prediction ability of the digital twin model. In addition, the machine learning model can also optimize the structure of the digital twin model. For example, by analyzing the complex relationships and patterns in the data, it is recommended to add or reduce certain layers, change the connection method, etc., to make the structure of the digital twin model more reasonable and efficient. Finally, apply the adjusted parameters and optimized structure to the digital twin model to generate an updated digital twin model, and use an independent test data set to verify and evaluate the updated digital twin model to check whether the performance of the digital twin model meets the expected requirements. If the test results are not satisfactory, it may be necessary to return to the data collection or machine learning model training stage to further adjust and optimize the process.

[0039] S3. Construct a reinforcement learning environment based on the trained digital twin model, define a reinforcement learning agent, and generate an optimal control strategy; Based on the updated digital twin model, analyze the operation data and dynamic characteristics of the production process, and define the state space and action space; Furthermore, first, extract the operation data of the production process from the updated digital twin model, including equipment status monitoring data, production plan execution data, material flow data, etc. At the same time, collect the real-time data in the physical entity production process and perform preprocessing operations such as cleaning and normalization to ensure the quality and consistency of the data. Then, analyze the dynamic characteristics of the production process through methods such as time series analysis, correlation analysis, and fault analysis, identify the factors and variables affecting the production process, such as changes in the operating status of equipment, adjustments to production plans, fluctuations in material supply and demand, etc., and reveal the internal laws and dynamic behavior patterns of the production process through data analysis and modeling. Next, according to the internal laws and dynamic behavior patterns of the production process obtained from the analysis, define the dimension and value range of the state space. The state space should be able to comprehensively describe the state characteristics of the production process, including equipment status, production progress, material inventory, etc. At the same time, combined with the control requirements and operation specifications of the production process, define the dimension and value range of the action space. The action space should cover all possible control actions, such as operations like starting, stopping, and adjusting parameters of equipment; decisions such as changing production plans and task allocation; measures such as replenishing and scheduling materials. Finally, according to specific business requirements and control objectives, optimize and adjust the state space and action space to ensure that their definitions can accurately reflect the characteristics and requirements of the production process and have good operability and controllability. For example, according to actual production requirements, adjust and optimize the monitoring frequency of state variables, the execution accuracy of actions, etc.

[0040] Based on historical production data and performance indicators, generate optimization objectives through the analysis of the digital twin model, and define the reward function in combination with dynamic characteristics; It should be noted that the optimization objectives specifically refer to identifying the bottlenecks and improvement points in the production process by analyzing historical production data and performance indicators, so as to determine the key indicators and directions that need to be optimized. These optimization objectives can include multiple aspects such as improving production efficiency, reducing equipment failure rates, reducing material waste, shortening the production cycle, and improving product quality. For example, improving production efficiency can be broken down into specific objectives such as reducing equipment idle time and optimizing production plan scheduling; reducing equipment failure rates involves improving the reliability and maintenance efficiency of equipment. The establishment of optimization objectives is the foundation of the entire process. It guides the definition of the subsequent reward function and the optimization direction of the production process. The ultimate goal is to make the production process more efficient, stable, and sustainable through the analysis and optimization of the digital twin model.

[0041] The specific process is as follows. First, use the digital twin model to deeply analyze historical production data and performance indicators, identify key problems and improvement points in the production process, and thus generate specific optimization goals, such as improving production efficiency, reducing equipment failure rates, etc. Then, analyze the dynamic characteristics of the production process, including changes in equipment operating status, adjustments to production plans, and fluctuations in material supply and demand, etc., to understand how these factors affect production. On this basis, combine the optimization goals and dynamic characteristics to define the reward function, and guide the production process towards the optimization goals by quantifying the behaviors in the production process. For example, reward behaviors that improve efficiency and punish behaviors that lead to failures. Finally, optimize and adjust the optimization goals and reward function according to actual production requirements and control objectives to ensure their effectiveness and operability in actual applications.

[0042] Based on the state space, action space, and reward function, use simulation environment integration and interface development to construct a reinforcement learning environment.

[0043] Furthermore, first, define the state space, action space, and reward function, and clarify the key state variables, control actions, and behavior quantification methods in the production process. Then, develop a simulation environment based on the digital twin model to ensure that it can accurately simulate the production process and provide an interface for interacting with the reinforcement learning algorithm. Next, develop a standardized interface to achieve data exchange and control interaction between the reinforcement learning algorithm and the simulation environment, including functions such as state observation, action execution, and reward feedback. Finally, integrate the simulation environment with the reinforcement learning algorithm, conduct tests and verifications to ensure the stability and reliability of the entire reinforcement learning environment, and lay a foundation for subsequent reinforcement learning training and optimization.

[0044] Based on the reinforcement learning algorithm, use the basic fully connected network method to define the neural network architecture, and construct a policy network and a value network; The specific process is as follows. First, according to the dimensions of the state space and action space, define the sizes of the input layer and output layer of the neural network. Usually, the input layer is consistent with the dimension of the state space, and the output layer corresponds to the dimension of the action space. Then, define the hidden layer structure of the network, generally including one or two hidden layers, and use the ReLU activation function after each layer to increase the non-linear expression ability. Then, use the Xavier initialization method to initialize the network parameters to ensure a reasonable distribution of the initial weights and avoid the problem of gradient disappearance or explosion in the initial stage of training. Finally, train the policy network and value network through an optimization algorithm (such as Adam), calculate the gradient using the reward signal feedback from the environment, and update the network parameters, so that the actions output by the policy network can maximize the cumulative return, and at the same time the value network can accurately evaluate the value of the state, thus providing decision support for the reinforcement learning process.

[0045] Define a policy update mechanism through the Actor-Critic method based on the policy network and the value network; Combine the policy network, the value network, and the policy update mechanism to generate a reinforcement learning agent; Furthermore, first define the state space and the action space, and then define and initialize the policy network and the value network. During the interaction with the environment, collect experience data, use the experience data to update the value network to more accurately evaluate the state value, and then update the parameters of the policy network using the policy gradient method according to the evaluation result of the value network, so that the action output by the policy network can maximize the cumulative reward. Through continuous iterative optimization, the policy network and the value network gradually converge to a better state, and finally integrate the trained policy network and value network into an intelligent agent to form a reinforcement learning agent to achieve the expected optimization goal Interact and train through the reinforcement learning environment and the reinforcement learning agent to generate interaction data, and optimize the policy network through the policy update mechanism to generate an optimal control policy.

[0046] It should be noted that the interaction data is the information generated during the interaction between the reinforcement learning agent and the environment, and is used to train and optimize the agent's policy. It includes state, action, reward, new state, and termination flag. The interaction data is used to evaluate the effectiveness of the policy, helping the agent learn which actions can bring higher cumulative rewards in specific states, and adjusting and optimizing the behavior policy accordingly.

[0047] The specific process is as follows: First, initialize the reinforcement learning agent and the reinforcement learning environment; then, at each time step, the reinforcement learning agent selects an action according to the current policy, the reinforcement learning environment executes the action and returns the reward and the new state, and the data is stored in the buffer; then, use the buffer data to update the parameters of the value network and the policy network; finally, through continuous iterative optimization, the policy network gradually converges to the optimal state and can generate an optimal control policy according to the input state.

[0048] S4. Adjust the production parameters according to the optimal control policy, feedback the production data through the digital twin model, regularly verify the optimization effect, and iteratively update the digital twin model and the optimal control policy.

[0049] Based on the optimal control policy, dynamically adjust the operating data in the hydrogen fuel cell production process, and combine the digital twin model to real-time feedback the production data; The specific process is as follows. First, use the optimal control strategy to optimize key indicators such as production efficiency, equipment operation status, and material usage. During the production process, adjust the operating parameters of the hydrogen fuel cell in real time according to the optimal control strategy, such as equipment operating speed, temperature, pressure, etc., to ensure that the production process is always in the optimal state. At the same time, through the digital twin model, various data in the production process are simulated and fed back in real time, such as equipment status, production progress, material inventory, etc., providing data support and decision-making basis for dynamic adjustment. Finally, combine the real-time feedback data with the optimal control strategy to continuously optimize the production process, improve production efficiency and product quality, and reduce operating costs.

[0050] Regularly compare the actual production data with the optimization goals, evaluate the optimization effects of the optimal control strategy on production efficiency, energy consumption, and product quality, and generate verification results; The specific process is as follows. First, collect key performance indicator data from the production process and use the digital twin model to feed back the operating data in real time to ensure the accuracy and consistency of the data. Then, set optimization goals, such as improving production efficiency, reducing energy consumption, and enhancing product quality, and identify the parameters affecting production efficiency and product quality through methods such as descriptive statistical analysis, correlation analysis, and regression analysis. Next, simulate the actual production process in a virtual environment, verify the performance of the optimal control strategy, and continuously adjust and optimize the control strategy through on-site commissioning. Finally, record the optimized production efficiency, energy consumption, and product quality data, generate a detailed verification report, and establish a real-time monitoring and feedback mechanism to continuously track the energy efficiency performance and product quality in the production process to ensure the continuous effectiveness of the optimization strategy.

[0051] Based on the verification results, update the digital twin model and the optimal control strategy.

[0052] Furthermore, according to the actual production data and verification results, update the parameters of the digital twin model. Use methods such as recursive least squares method and gradient descent method to update the linear model. For the non-linear model, update the non-linear model parameters through optimal estimation or setting a specific loss function. According to the updated digital twin model and actual production data, adjust the optimal control strategy, and retrain the control strategy through reinforcement learning algorithm to further optimize the production process. Record the updated parameters of the digital twin model and the control strategy, generate a detailed update report, and establish a real-time monitoring and feedback mechanism to continuously track the energy efficiency performance and product quality in the production process to ensure the effectiveness of the updated digital twin model and strategy.

[0053] This embodiment also provides a hydrogen fuel cell production management system based on digital twin technology, including: a data acquisition module, a model construction module, a strategy generation module, and an optimization module; the data acquisition module is used to collect the production data of hydrogen fuel cells in real time, perform data preprocessing, and generate standardized data; the model construction module is used to construct a digital twin model based on the standardized data by using finite element analysis and computational fluid dynamics strategies, and train the digital twin model with historical data; the strategy generation module is used to construct a reinforcement learning environment based on the trained digital twin model, define a reinforcement learning agent, and generate an optimal control strategy; the optimization module is used to adjust the production parameters according to the optimal control strategy, feedback the production data through the digital twin model, regularly verify the optimization effect, and iteratively update the digital twin model and the optimal control strategy.

[0054] This embodiment also provides a computer device applicable to the case of a hydrogen fuel cell production management method based on digital twin technology, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the hydrogen fuel cell production management method based on digital twin technology proposed in the above embodiment.

[0055] This computer device can be a terminal, and this computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of this computer device is used to provide computing and control capabilities. The memory of this computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of this computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a carrier network, NFC (near field communication), or other technologies. The display screen of this computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of this computer device can be a touch layer covered on the display screen, or a button, a trackball, or a touchpad set on the shell of the computer device, or an external keyboard, a touchpad, or a mouse, etc.

[0056] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the method for hydrogen fuel cell production management based on digital twin technology proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, abbreviated as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, abbreviated as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, abbreviated as EPROM), programmable read-only memory (Programmable Red-Only Memory, abbreviated as PROM), read-only memory (Read-Only Memory, abbreviated as ROM), magnetic memory, flash memory, magnetic disk or optical disc.

[0057] In summary, through the digital twin model constructed by finite element analysis and computational fluid dynamics strategy, the present invention can accurately capture the mechanical and fluid characteristics in the production of hydrogen fuel cells, reflect the actual production situation, simulate physical phenomena, provide a basis for the construction of the reinforcement learning environment and the generation of the optimal control strategy, and contribute to production optimization; the reinforcement learning agent autonomously learns the optimal strategy in the simulated environment, and the generated control strategy can dynamically adjust the production parameters, improve efficiency, reduce energy consumption, improve quality, and achieve intelligent decision-making and control.

[0058] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A hydrogen fuel cell production management method based on digital twin technology, characterized in that: include, Collect hydrogen fuel cell production data in real time, perform data preprocessing, and generate standardized data; Based on standardized data, a digital twin model is constructed using finite element analysis and computational fluid dynamics strategies, and the digital twin model is trained using historical data; Build a reinforcement learning environment based on the trained digital twin model, define the reinforcement learning agent, and generate the optimal control strategy; Adjust production parameters according to the optimal control strategy, feedback production data through the digital twin model, regularly verify the optimization effect, and iteratively update the digital twin model and optimal control strategy.

2. The hydrogen fuel cell production management method based on digital twin technology according to claim 1, characterized in that: The production data includes temperature, pressure, gas flow, current density, equipment status, quality test results and operation logs; The preprocessing is to clean, standardize, extract features and reduce dimension of the data to generate standardized data.

3. The hydrogen fuel cell production management method based on digital twin technology according to claim 2, characterized in that: The digital twin model is constructed based on standardized data using finite element analysis and computational fluid dynamics strategies. The specific steps are as follows: Based on production data, define the input, output, and internal structure of the digital twin model and generate the architecture of the initial digital twin model; Based on the standardized data and the architecture of the initial digital twin model, a multi-physics model is constructed according to the structural mechanics and fluid dynamics problems in the production process; Through the meshing method, the multi-physics field model is divided to generate finite element mesh and fluid dynamics mesh; Through the coupling algorithm, the finite element analysis grid and the computational fluid dynamics grid are combined to generate a digital twin model.

4. The hydrogen fuel cell production management method based on digital twin technology according to claim 3, characterized in that: The specific steps of training the digital twin model through historical data are as follows: Divide historical data into training set, validation set and test set; The digital twin model is initialized using the initialization method Xavier, the digital twin model is trained using the training set, the hyperparameters of the digital twin model are adjusted using the validation set, the performance of the digital twin model is evaluated using the test set, and the trained digital twin model is generated; Through the machine learning model, the parameters and structure of the digital twin model are dynamically adjusted to generate an updated digital twin model.

5. The hydrogen fuel cell production management method based on digital twin technology according to claim 4, characterized in that: The specific steps of building a reinforcement learning environment based on the trained digital twin model are as follows: Based on the updated digital twin model, analyze the operating data and dynamic characteristics of the production process and define the state space and action space; Based on historical production data and performance indicators, the optimization objectives are generated through the analysis of the digital twin model, and the reward function is defined in combination with dynamic characteristics; Based on the state space, action space and reward function, a reinforcement learning environment is constructed by using simulation environment integration and interface development.

6. The hydrogen fuel cell production management method based on digital twin technology according to claim 5, characterized in that: The steps of defining the reinforcement learning agent and generating the optimal control strategy are as follows: Based on the reinforcement learning algorithm, the basic fully connected network method is used to define the neural network architecture and build the strategy network and value network; Based on the policy network and value network, the policy update mechanism is defined through the Actor-Critic method; Combine the policy network, value network, and policy update mechanism to generate a reinforcement learning agent; Through interactive training between the reinforcement learning environment and the reinforcement learning agent, interactive data is generated, and the policy network is optimized through the policy update mechanism to generate the optimal control strategy.

7. The hydrogen fuel cell production management method based on digital twin technology according to claim 6, characterized in that: The specific steps of adjusting production parameters according to the optimal control strategy, feeding back production data through the digital twin model, regularly verifying the optimization effect, and iteratively updating the digital twin model and the optimal control strategy are as follows: Based on the optimal control strategy, the equipment parameters in the hydrogen fuel cell production process are dynamically adjusted, and the production data is fed back in real time in combination with the digital twin model; Regularly compare actual production data with optimization targets, evaluate the optimization effect of the optimal control strategy on production efficiency, energy consumption and product quality, and generate verification results; Based on the verification results, the digital twin model and optimal control strategy are updated.

8. A hydrogen fuel cell production management system based on digital twin technology, based on the hydrogen fuel cell production management method based on digital twin technology according to any one of claims 1 to 7, characterized in that: Including data acquisition module, model building module, strategy generation module and optimization module; Data acquisition module, used to collect production data of hydrogen fuel cells in real time, perform data preprocessing, and generate standardized data; A model building module, which is used to build a digital twin model based on standardized data using finite element analysis and computational fluid dynamics strategies, and train the digital twin model with historical data; The strategy generation module is used to build a reinforcement learning environment based on the trained digital twin model, define the reinforcement learning agent, and generate the optimal control strategy; The optimization module is used to adjust production parameters according to the optimal control strategy, feedback production data through the digital twin model, regularly verify the optimization effect, and iteratively update the digital twin model and optimal control strategy.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the hydrogen fuel cell production management method based on digital twin technology described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the hydrogen fuel cell production management method based on digital twin technology described in any one of claims 1 to 7 are implemented.

Images