Digital twinning-based AEM hydrogen production galvanic pile cooperative control system and method

Abstract

The invention relates to the technical field of electrolytic hydrogen production, in particular to an AEM hydrogen production galvanic pile cooperative control system and method based on digital twinning, and the system comprises an equipment layer which comprises a plurality of galvanic pile clusters, and each galvanic pile cluster comprises a plurality of AEM hydrogen production galvanic piles; the cloud platform layer comprises a digital twinborn body, and the digital twinborn body integrates an electrochemical model, a fluid dynamic model and a historical degradation data processing unit and is used for receiving the original measurement value, predicting the performance attenuation trend of each electric pile and generating a prediction result quantified as a health index; the control layer comprises a centralized controller and a plurality of cluster control units; the centralized controller comprises a fault analysis module and a global load distribution module; the fault analysis module is used for analyzing the fault mark to determine the fault cluster position and the load gap, and the global load distribution module is used for calculating the load compensation amount of the health cluster, generating a global load distribution instruction and issuing the global load distribution instruction to the health cluster.

Description

Technical Field

[0001] The present invention relates to the technical field of electrolytic hydrogen production, and in particular to a collaborative control system and method for an AEM hydrogen production stack based on digital twins. Background Art

[0002] AEM (anion exchange membrane) hydrogen production technology has become a research hotspot in the field of water electrolysis hydrogen production in recent years due to its high efficiency, low energy consumption and environmental friendliness. However, AEM fuel cells face many challenges in actual operation: on the one hand, the performance of the fuel cell will decay over time due to aging of the membrane material, deactivation of the catalyst or degradation of the fluid dynamic characteristics, resulting in a decrease in hydrogen production efficiency; on the other hand, the AEM fuel cell is highly sensitive to operating conditions (such as voltage fluctuations, temperature gradients, and gas pressure), and traditional control strategies are difficult to achieve accurate distribution of dynamic loads and rapid response to faults. Existing control systems mostly adopt a centralized architecture and lack the ability to adapt to differences between fuel cell clusters, making it difficult to optimize the overall system stability and energy efficiency. Therefore, there is an urgent need for a collaborative control solution that can integrate real-time data, predictive models and intelligent decision-making to improve the operating efficiency of the AEM fuel cell, extend its life and ensure system reliability. Summary of the Invention

[0003] In order to solve the problems in the related art, the embodiments of the present disclosure provide an AEM hydrogen production stack collaborative control system and method based on digital twin.

[0004] In a first aspect, the present disclosure provides an AEM hydrogen production stack collaborative control system based on digital twins, including:

[0005] The equipment layer includes multiple fuel cell clusters, each of which contains multiple AEM hydrogen production fuel cells. Each AEM hydrogen production fuel cell is equipped with a sensor group and an actuator group. The sensor group is used to collect the raw measurement values ​​of the AEM hydrogen production fuel cell in real time, and the actuator group is used to implement the load distribution plan within the cluster.

[0006] The cloud platform layer includes a digital twin that integrates an electrochemical model, a fluid dynamics model, and a historical degradation data processing unit to receive the raw measurements and predict the performance degradation trend of each fuel cell stack, generating a prediction result quantified as a health index.

[0007] The control layer includes a centralized controller and multiple cluster control units;

[0008] Each cluster control unit includes a real-time operation feature calculation module, a local dynamic load optimization model, and a data synchronization mechanism; the real-time operation feature calculation module is used to calculate the real-time operation feature data based on the received raw measurement values; the local dynamic load optimization model generates the load distribution plan within the cluster based on the reinforcement learning algorithm, the health index, and the real-time operation feature data; the data synchronization mechanism is used to calculate the total cluster output power and monitor anomalies to generate fault flags, and only transmit the total cluster output power and fault flags to the centralized controller;

[0009] The centralized controller includes a fault analysis module and a global load distribution module; the fault analysis module is used to analyze the fault sign to determine the location of the fault cluster and the load gap; the global load distribution module is used to calculate the load compensation amount of the healthy cluster, generate a global load distribution instruction and send it to the cluster control unit of the healthy cluster, so that the actuator group of the healthy cluster executes the cluster load distribution plan to maintain the stability of the total output power of the cluster.

[0010] According to an embodiment of the present disclosure, the health index Hi generated by the digital twin is calculated by the following formula:

[0011]

[0012] Where ΔVi is the voltage offset, Vnom is the rated voltage, Ti is the real-time temperature, and top is the cumulative operating hours.

[0013] According to an embodiment of the present disclosure, the data packet transmitted by the data synchronization mechanism consists of 5 bytes, wherein the first 4 bytes represent the single-precision floating-point number of the total cluster output power, and the 5th byte represents the fault flag, which includes normal state, warning state and serious fault state.

[0014] According to an embodiment of the present disclosure, it also includes: a dynamic load distribution module, which is used to generate a load distribution strategy based on the health index generated by the digital twin.

[0015] According to an embodiment of the present disclosure, the load distribution strategy is as follows:

[0016] When the health index Hi is greater than or equal to 0.8, 95% to 100% of the baseline load is assigned;

[0017] When the health index Hi is greater than or equal to 0.6 and less than 0.8, 70% to 95% of the baseline load is assigned;

[0018] When the health index Hi is less than 0.6, no more than 60 percent of the baseline load is allocated.

[0019] According to an embodiment of the present disclosure, the part of the global load distribution module that calculates the load compensation amount of the healthy cluster, generates the global load distribution instruction, and sends it to the cluster control unit of the healthy cluster is configured as follows:

[0020] Eliminate faulty clusters and identify all healthy clusters;

[0021] For each health cluster, the amount of load compensation that should be increased is calculated according to the proportion of its health index in the total health index;

[0022] The calculated load compensation amount is distributed to the cluster control units of the corresponding healthy clusters.

[0023] According to an embodiment of the present disclosure, each cluster control unit also includes a lightweight AI reasoning module for making millisecond-level decisions based on the real-time operating characteristic data to generate single-stack emergency control instructions.

[0024] According to an embodiment of the present disclosure, the lightweight AI reasoning module deploys and runs a quantized deep Q network model;

[0025] The model size of the quantized deep Q network model is no more than 8 megabytes, and the inference delay is no more than 8 milliseconds.

[0026] The reward function of the quantized deep Q network model is a times the hydrogen production efficiency minus b times the voltage fluctuation minus c times the number of overtemperatures; where the sum of the values ​​of a, b, and c is 1.

[0027] In a second aspect, the present disclosure provides a collaborative control method for an AEM hydrogen production stack based on digital twins, comprising the following steps:

[0028] S1: The sensor group at the equipment layer collects the raw measurement values ​​of voltage, temperature, and pressure of each AEM hydrogen production stack in real time, and synchronously uploads the raw measurement values ​​to the cloud platform layer and the cluster control unit at the control layer;

[0029] S2: The digital twin at the cloud platform layer predicts the performance degradation trend of each fuel cell stack based on the raw measurement values, coupled with the electrochemical model, fluid dynamics model, and historical degradation data. The prediction results are quantified as a fuel cell stack health index, and the health index is sent to the centralized controller and cluster control unit at the control layer.

[0030] S3: The cluster control unit calculates real-time operating characteristic data based on the received raw measurement values, including voltage fluctuation rate, temperature gradient, and pressure change trend. It uses a lightweight AI inference module to make millisecond-level decisions based on the real-time operating characteristic data to generate single-stack emergency control instructions. It also generates a cluster load distribution plan based on the reinforcement learning algorithm, health index, and real-time operating characteristic data through a local dynamic load optimization model.

[0031] S4: The cluster control unit calculates the total cluster output power through a data synchronization mechanism, monitors for anomalies, and generates a fault flag. It only transmits the total cluster output power and the fault flag to the centralized controller. When the fault flag indicates that cross-cluster coordination is required, it triggers the generation of a global load distribution instruction.

[0032] S5: The centralized controller analyzes the fault flag to determine the location of the faulty cluster and the load shortfall, calculates the load compensation amount for the healthy cluster, which is equal to the load shortfall multiplied by the ratio of the healthy cluster's health index to the total health index, generates a global load allocation instruction, and sends it to the cluster control units of the healthy clusters, causing them to increase their hydrogen production load according to the compensation amount to maintain the stability of the system's total output power.

[0033] S6: The actuator group at the device layer executes the load distribution plan within the cluster, responds to the emergency control instructions of the single stack, and collects a new round of raw measurement values ​​to form a closed-loop control flow.

[0034] In a third aspect, an embodiment of the present disclosure provides an electronic device comprising a memory and a processor, wherein the memory is used to store one or more computer instructions, and wherein the one or more computer instructions are executed by the processor to implement the method described in the second aspect.

[0035] In a fourth aspect, an embodiment of the present disclosure provides a computer-readable storage medium on which computer instructions are stored. When the computer instructions are executed by a processor, the method described in the second aspect is implemented.

[0036] The technical effects provided by the embodiments of the present disclosure may include the following beneficial effects:

[0037] According to the technical solution provided by the embodiment of the present disclosure, the AEM hydrogen production stack collaborative control system based on digital twins includes: an equipment layer, including multiple stack clusters, each stack cluster includes multiple AEM hydrogen production stacks; each AEM hydrogen production stack is equipped with a sensor group and an actuator group, the sensor group is used to collect the original measurement values ​​of the AEM hydrogen production stack in real time, and the actuator group is used to execute the load distribution plan within the cluster; a cloud platform layer, including a digital twin, the digital twin integrates an electrochemical model, a fluid dynamics model and a historical degradation data processing unit, and is used to receive the original measurement values ​​and predict the performance attenuation trend of each stack, and generate a prediction result quantified as a health index; a control layer, including a centralized controller and multiple cluster control units; wherein each cluster control unit includes a real-time operation feature calculation module, a local dynamic load optimization model and a data synchronization machine system; the real-time operation feature calculation module is used to calculate the real-time operation feature data based on the received original measurement value; the local dynamic load optimization model generates the intra-cluster load distribution plan based on the reinforcement learning algorithm, the health index and the real-time operation feature data; the data synchronization mechanism is used to calculate the total cluster output power and monitor anomalies to generate fault signs, and only transmit the total cluster output power and fault signs to the centralized controller; the centralized controller includes a fault analysis module and a global load distribution module; the fault analysis module is used to analyze the fault sign to determine the fault cluster location and load gap, and the global load distribution module is used to calculate the load compensation amount of the healthy cluster, generate a global load distribution instruction and send it to the cluster control unit of the healthy cluster, so that the actuator group of the healthy cluster executes the intra-cluster load distribution plan to maintain the stability of the total cluster output power.

[0038] The above technical solution achieves full lifecycle management of the AEM stack's operating status through a layered architecture consisting of the device layer, cloud platform layer, and control layer. The sensor and actuator groups at the device layer provide the system with high-precision real-time monitoring and rapid response capabilities, addressing the issue of insufficient dynamic adjustment in traditional control systems due to data acquisition delays or execution lags. The digital twin at the cloud platform layer integrates electrochemical models, fluid dynamics models, and a historical degradation data processing unit to construct a prediction system for stack performance degradation. This allows the generation of a health index that combines the advantages of both physical mechanisms and data-driven approaches. This prediction provides a scientific basis for dynamic load optimization at the control layer, avoiding blind loading or overly conservative operating strategies. The cluster control unit at the control layer achieves multi-scale load distribution from a single stack to a cluster of stacks by collaborating with a local dynamic load optimization model (based on a reinforcement learning algorithm) and a global load distribution module. The reinforcement learning algorithm dynamically adjusts the load distribution strategy based on the health index and real-time operating characteristics, maximizing hydrogen production efficiency while ensuring stack safety. The data synchronization mechanism uses a lightweight protocol to transmit only critical information (cluster total power and fault indicators), significantly reducing communication overhead and shortening response time while ensuring the centralized controller's real-time awareness of global status. When a fault occurs, the fault analysis module, in conjunction with the global load distribution module, quickly locates the source and compensates healthy clusters for load, effectively mitigating the impact of single-point failures on system output.

[0039] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 An architectural diagram of the AEM hydrogen production stack collaborative control system based on digital twin according to an embodiment of the present disclosure is shown.

[0041] Figure 2 A flow chart of a collaborative control method of an AEM hydrogen production stack based on digital twin according to an embodiment of the present disclosure is shown.

[0042] Figure 3 A structural block diagram of an electronic device according to an embodiment of the present disclosure is shown.

[0043] Figure 4 A schematic diagram showing the structure of a computer system suitable for implementing the method according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0044] Hereinafter, exemplary embodiments of the present disclosure will be described in detail with reference to the accompanying drawings so that those skilled in the art can easily implement them. In addition, for the sake of clarity, parts not related to the description of the exemplary embodiments are omitted in the accompanying drawings.

[0045] In the present disclosure, it should be understood that terms such as "include" or "have" are intended to indicate the presence of features, numbers, steps, actions, components, parts, or combinations thereof disclosed in the present specification, and are not intended to exclude the possibility that one or more other features, numbers, steps, actions, components, parts, or combinations thereof exist or are added.

[0046] It should also be noted that, in the absence of conflict, the embodiments and features of the embodiments of the present disclosure may be combined with each other. The present disclosure will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0047] AEM (anion exchange membrane) hydrogen production technology has become a research hotspot in the field of water electrolysis hydrogen production in recent years due to its high efficiency, low energy consumption and environmental friendliness. However, AEM fuel cells face many challenges in actual operation: on the one hand, the performance of the fuel cell will decay over time due to aging of the membrane material, deactivation of the catalyst or degradation of the fluid dynamic characteristics, resulting in a decrease in hydrogen production efficiency; on the other hand, the AEM fuel cell is highly sensitive to operating conditions (such as voltage fluctuations, temperature gradients, and gas pressure), and traditional control strategies are difficult to achieve accurate distribution of dynamic loads and rapid response to faults. Existing control systems mostly adopt a centralized architecture and lack the ability to adapt to differences between fuel cell clusters, making it difficult to optimize the overall system stability and energy efficiency. Therefore, there is an urgent need for a collaborative control solution that can integrate real-time data, predictive models and intelligent decision-making to improve the operating efficiency of the AEM fuel cell, extend its life and ensure system reliability.

[0048] Taking the above-mentioned defects into consideration, the AEM hydrogen production stack collaborative control system based on digital twins provided in the embodiment of the present disclosure realizes the full life cycle management of the operating status of the AEM stack. The sensor group and actuator group at the equipment layer provide the system with high-precision real-time monitoring and rapid response capabilities, solving the problem of insufficient dynamic adjustment caused by data acquisition delays or execution lags in traditional control. The digital twin at the cloud platform layer integrates electrochemical models, fluid dynamics models and historical degradation data processing units to build a prediction system for stack performance degradation, so that the generation of health index has the dual advantages of physical mechanism and data drive. The prediction results provide a scientific basis for the dynamic load optimization of the control layer, avoiding blind loading or overly conservative operation strategies. The cluster control unit of the control layer realizes multi-scale load distribution from single stack to stack cluster through the collaboration of local dynamic load optimization model and global load distribution module. The reinforcement learning algorithm can dynamically adjust the load distribution strategy according to the health index and real-time operation characteristics, maximizing the hydrogen production efficiency while ensuring the safety of the stack. The data synchronization mechanism uses a lightweight protocol to transmit only critical information, significantly reducing communication overhead and shortening response time, while also ensuring the centralized controller's real-time awareness of global status. When a fault occurs, the fault analysis module, in conjunction with the global load distribution module, quickly locates the source and compensates the load of healthy clusters, effectively mitigating the impact of single-point failures on system output.

[0049] Figure 1 An architectural diagram of the AEM hydrogen production stack collaborative control system based on digital twin according to an embodiment of the present disclosure is shown.

[0050] like Figure 1 As shown in the figure, the AEM hydrogen production stack collaborative control system based on digital twin includes the equipment layer, cloud platform layer and control layer.

[0051] The equipment layer includes multiple fuel cell clusters, such as fuel cell cluster 1, fuel cell cluster 2...fuel cell cluster n, each fuel cell cluster contains multiple AEM hydrogen production fuel cells, such as fuel cell cluster 1 includes AEM fuel cell 1, AEM fuel cell 2...AEM fuel cell n, where n is a natural number; each AEM hydrogen production fuel cell is equipped with a sensor group and an actuator group, the sensor group is used to collect the original measurement values ​​of the AEM hydrogen production fuel cell in real time, and the actuator group is used to execute the load distribution plan within the cluster.

[0052] Specifically, the sensor group includes a voltage sensor (for measuring single stack or total voltage), a temperature sensor (for monitoring the temperature of the electrolyte or the stack body), a pressure sensor (for detecting the hydrogen / oxygen outlet pressure), a current sensor (for collecting working current), etc. For example, the sampling frequency of the voltage sensor is 1kHz, the accuracy of the temperature sensor is ±0.1°C, and the range of the pressure sensor is 0-2MPa. The sensor group collects the original measurement values ​​of the AEM stack in real time, such as voltage fluctuations, temperature distribution, gas pressure, and current intensity. The original measurement values ​​collected by the sensor group can also be transmitted to the cloud platform layer after preprocessing (such as filtering and denoising). For example, the voltage signal may contain high-frequency noise, and the effective value needs to be extracted through a low-pass filter (cut-off frequency 100Hz); the temperature data needs to be calibrated for sensor deviation (such as correcting zero drift through a calibration curve).

[0053] The actuator group includes current regulators (such as DC-DC converters), cooling system valves (to adjust heat dissipation efficiency), and gas flow control valves (to adjust hydrogen and oxygen separation efficiency). For example, the current regulator has a response time of 5ms, and the cooling system valve has an opening adjustment range of 0-100%. The actuator group adjusts the operating parameters of the fuel cell stack based on the load distribution plan within the cluster to ensure output power and safe operation.

[0054] The cloud platform layer includes a digital twin, which integrates an electrochemical model, a fluid dynamics model and a historical degradation data processing unit to receive the original measurement values ​​and predict the performance degradation trend of each battery stack, and generate a prediction result quantified as a health index.

[0055] Specifically, the electrochemical model can simulate the kinetics of the hydrolysis reaction based on the Nernst equation and the Butler-Volmer equation. For example, the rate equation for the cathode hydrogen evolution reaction (HER) is:

[0056]

[0057] Where j0 is the exchange current density, α is the transfer coefficient, and η is the overpotential.

[0058] The fluid dynamics model can use the Navier-Stokes equations to simulate the electrolyte flow and calculate the flow rate, pressure gradient and mass transfer efficiency. For example, the permeability of the electrolyte in the porous electrode needs to be calculated by Darcy's law.

[0059] quantized, where k is the permeability and μ is the viscosity.

[0060] The historical degradation data processing unit can analyze long-term operating data (such as voltage decay records over a 10-year period) based on an LSTM neural network to predict stack performance degradation trends. For example, time series analysis can identify stack aging patterns, such as exponential growth in membrane resistance over time.

[0061] The digital twin is a virtual mirror of a physical entity, enabling full lifecycle management of the AEM stack through real-time data updates and simulation predictions. Its core features include multi-physics coupled modeling, such as electrochemical and fluid dynamics models, bidirectional data exchange (e.g., physical entity → virtual model → physical entity), and predictive maintenance capabilities.

[0062] Health Index H i Refers to the indicator that quantifies the performance degradation of the battery stack, ranging from 0 to 1. For example, H i ≥0.8 indicates that the battery stack is healthy, H i <0.6 indicates that the machine needs to be shut down for maintenance.

[0063] Raw measurements (such as voltage and temperature) are pre-processed by edge computing nodes and uploaded to the cloud platform. The digital twin combines this with model predictions to generate a health index. For example, voltage offsets are calculated using a sliding window mean filter, and temperature data is smoothed using a Kalman filter to eliminate abnormal fluctuations.

[0064] The control layer includes a centralized controller and multiple cluster control units, such as cluster control unit 1, cluster control unit 2, ..., cluster control unit n, where n is a natural number and each cluster control unit corresponds to a battery stack cluster;

[0065] Among them, each cluster control unit includes a real-time operation feature calculation module, a local dynamic load optimization model and a data synchronization mechanism; the real-time operation feature calculation module is used to calculate the real-time operation feature data based on the received original measurement value; the local dynamic load optimization model generates the load distribution plan within the cluster based on the reinforcement learning algorithm, the health index and the real-time operation feature data; the data synchronization mechanism is used to calculate the total cluster output power and monitor anomalies to generate fault signs, and only transmit the total cluster output power and fault signs to the centralized controller.

[0066] Specifically, the real-time operation characteristic calculation module calculates the real-time operation characteristics of the fuel cell stack, such as voltage fluctuation rate and temperature gradient, based on sensor data.

[0067] The local dynamic load optimization model uses a reinforcement learning algorithm to generate a load distribution plan within the cluster. The reinforcement learning algorithm in this disclosure is a DDPG (Deep Deterministic Policy Gradient) algorithm combined with an Actor-Critic framework. The Actor network generates load distribution actions, and the Critic network evaluates the Q value of the action (i.e., the expected value of the reward function R). The model input includes the health index H iAnd real-time characteristics (such as voltage fluctuation rate), output the specific allocation ratio of each battery stack. Among them, the reward function R is:

[0068] R=a·η-b·ΔV-c·N overtemp ;

[0069] Among them, η is the hydrogen production efficiency, ΔV is the voltage fluctuation, N overtemp is the number of overtemperatures. The value of a can be 0.7, the value of b can be 0.2, and the value of c can be 0.1.

[0070] The cluster control unit calculates the total cluster output power and generates a fault flag (0x00 normal, 0x01 warning, 0x02 severe fault). For example, if a stack temperature exceeds the limit (>80°C), the flag is set to 0x01 and uploaded to the centralized controller via a custom protocol (5-byte packet).

[0071] The cluster control unit's data synchronization mechanism uses a lightweight protocol that transmits only the cluster total power (a single-precision floating-point number, 4 bytes) and a fault flag (1 byte), reducing the communication load to less than 100 kbps. For example, the IEEE 754 code for a cluster total power of 350 kW is 0x43333333, and the fault flag 0x01 indicates a warning state.

[0072] The centralized controller includes a fault analysis module and a global load distribution module; the fault analysis module is used to analyze the fault sign to determine the location of the fault cluster and the load gap; the global load distribution module is used to calculate the load compensation amount of the healthy cluster, generate a global load distribution instruction and send it to the cluster control unit of the healthy cluster, so that the actuator group of the healthy cluster executes the cluster load distribution plan to maintain the stability of the total output power of the cluster.

[0073] Specifically, the fault analysis module analyzes the fault flags to locate the fault cluster. For example, if the 0x02 flag is received, the controller invokes a fault tree analysis (FTA) to determine the source of the fault (such as membrane rupture or catalyst failure) and calculates the load gap ΔP.

[0074] The global load distribution module distributes the compensation amount according to the health index ratio of the healthy cluster. For example, if there are two healthy clusters with health indexes H1 = 0.85 and H2 = 0.75, the compensation amount is and distribute.

[0075] According to an embodiment of the present disclosure, the health index H generated by the digital twin i Calculated by the following formula:

[0076]

[0077] Where, ΔVi is the voltage offset, V nom Indicates rated voltage, T i is the real-time temperature, and top is the cumulative running hours.

[0078] For example, the ΔVi of a certain battery stack is 0.3V, V nom 2.0V, T i is 65℃, top is 10000h, then H i It is 0.75, indicating that the stack is in a healthy state but needs to be operated at a reduced load.

[0079] According to an embodiment of the present disclosure, the data packet transmitted by the data synchronization mechanism consists of 5 bytes, wherein the first 4 bytes represent the single-precision floating-point number of the total cluster output power, and the 5th byte represents the fault flag, which includes normal state, warning state and serious fault state.

[0080] Specifically, the single-precision floating-point number of the cluster total output power may adopt the IEEE 754 format. For example, the binary representation corresponding to 350 kW is 0x43333333.

[0081] The fault flag uses 8-bit binary and is defined as follows:

[0082] 0x00 indicates normal status (no abnormality);

[0083] 0x01 indicates a warning state (such as a slight temperature exceeding a limit or a voltage fluctuation exceeding a threshold);

[0084] 0x02 identifies a serious fault condition (such as membrane rupture, current overload).

[0085] The cluster control unit periodically calculates the total cluster output power, for example, every 100 milliseconds, and generates a fault flag by sampling sensor data using the ADC. Data packets are transmitted to the centralized controller via industrial Ethernet (such as Profinet), using the custom frame format described above to reduce bandwidth usage.

[0086] According to an embodiment of the present disclosure, it also includes: a dynamic load distribution module, which is used to generate a load distribution strategy based on the health index generated by the digital twin.

[0087] Specifically, after the digital twin generates Hi, the dynamic load distribution module adjusts the stack current through the PID controller to ensure that the output power matches the load strategy.

[0088] According to an embodiment of the present disclosure, the load distribution strategy is as follows:

[0089] When the health index Hi is greater than or equal to 0.8, 95% to 100% of the baseline load is assigned;

[0090] When the health index Hi is greater than or equal to 0.6 and less than 0.8, 70% to 95% of the baseline load is assigned;

[0091] When the health index Hi is less than 0.6, no more than 60 percent of the baseline load is allocated.

[0092] According to an embodiment of the present disclosure, the part of the global load distribution module that calculates the load compensation amount of the healthy cluster, generates the global load distribution instruction, and sends it to the cluster control unit of the healthy cluster is configured as follows:

[0093] Eliminate faulty clusters and identify all healthy clusters;

[0094] For each health cluster, the amount of load compensation that should be increased is calculated according to the proportion of its health index in the total health index;

[0095] The calculated load compensation amount is distributed to the cluster control units of the corresponding healthy clusters.

[0096] Specifically, when a cluster failure occurs, the global load distribution module performs the following steps:

[0097] Parse the fault flag bit (e.g., 0x02), mark the faulty cluster, and stop powering it.

[0098] Calculate the load gap. If the fault cluster originally contributed 100kW, the load gap is 100kW.

[0099] The compensation amount is allocated according to the health index ratio. For example, if H1 = 0.85 and H2 = 0.75, the compensation amounts are:

[0100] Cluster 1: ΔP × (0.85) / (0.85 + 0.75) = 78.8 kW;

[0101] Cluster 2: ΔP × (0.75) / (0.85 + 0.75) = 61.2 kW;

[0102] The command is issued to convert the compensation amount into a current increment and sent to the actuator group of the healthy cluster via the CAN bus.

[0103] According to an embodiment of the present disclosure, each cluster control unit also includes a lightweight AI reasoning module for making millisecond-level decisions based on the real-time operating characteristic data to generate single-stack emergency control instructions.

[0104] Specifically, the lightweight AI inference module's model structure consists of an input layer, hidden layers, and an output layer. The input layer inputs real-time operational characteristics such as voltage fluctuation and temperature gradient. The hidden layers consist of three fully connected layers, each with 64 neurons and a Swish activation function. The output layer outputs emergency control commands for the single stack, such as current adjustment steps of ±5% and cooling valve opening adjustment.

[0105] According to an embodiment of the present disclosure, the lightweight AI reasoning module deploys and runs a quantized deep Q network model;

[0106] The model size of the quantized deep Q network model is no more than 8 megabytes, and the inference delay is no more than 8 milliseconds.

[0107] The reward function of the quantized deep Q network model is a times the hydrogen production efficiency minus b times the voltage fluctuation minus c times the number of overtemperatures; where the sum of the values ​​of a, b, and c is 1.

[0108] Specifically, model parameters are quantized to 8-bit integers, compressing the model size to 7.8MB. Inference latency is kept within 8ms by removing redundant connections through pruning and hardware acceleration, such as FPGAs.

[0109] The reward function design is consistent with the local dynamic load optimization model. Both reward functions use hydrogen production efficiency as the core metric and penalize voltage fluctuations and overtemperature, reflecting a unified safety priority. The local dynamic load optimization model (DDPG) is responsible for macro-level scheduling, such as cluster-level load distribution, while the quantitative deep Q network model (DQN) is responsible for micro-level emergency response, such as single-stack regulation. The two complement each other through different optimization objectives.

[0110] Application scenario examples:

[0111] An integrated energy station deployed an AEM hydrogen production system (total power of 500 kW, consisting of 10 fuel cell clusters, each containing five AEM fuel cells). The system needed to cope with challenges such as power fluctuations, fuel cell aging, and sudden failures to achieve efficient, stable, and safe hydrogen production.

[0112] Traditional control scheme

[0113] A fixed load distribution strategy is adopted, and each fuel cell stack operates at a baseline power without considering changes in health status or operating conditions.

[0114] The voltage, current and other parameters are adjusted by the PID controller, and the response delay is about 500ms.

[0115] Troubleshooting requires manual troubleshooting and long downtime (e.g., membrane rupture requires 30 minutes of downtime).

[0116] This application control plan

[0117] Digital twin prediction: The cloud platform layer generates a stack health index through electrochemical models, fluid dynamics models, and historical degradation data to drive dynamic load distribution.

[0118] Hierarchical collaborative control:

[0119] Cluster control unit: Generates a load distribution plan within the cluster based on the stack health index and real-time operating characteristics.

[0120] Centralized controller: Analyzes fault signs, calculates the compensation amount for healthy clusters, and issues global load instructions.

[0121] Lightweight AI reasoning: Quantized deep Q network models enable single-stack emergency control to suppress voltage fluctuations and overheating.

[0122] The comparison between the traditional control scheme and the control scheme of this application in terms of hydrogen production efficiency, power stability, emergency response time, stack life management and fault handling capability is shown in the following table:

[0123]

[0124] Figure 2 A flow chart of a collaborative control method of an AEM hydrogen production stack based on digital twin according to an embodiment of the present disclosure is shown.

[0125] like Figure 2 As shown in FIG, the collaborative control method of the AEM hydrogen production stack based on digital twin includes the following steps:

[0126] S1: The sensor group at the equipment layer collects the raw measurement values ​​of voltage, temperature, and pressure of each AEM hydrogen production stack in real time, and synchronously uploads the raw measurement values ​​to the cloud platform layer and the cluster control unit at the control layer;

[0127] S2: The digital twin at the cloud platform layer predicts the performance degradation trend of each fuel cell stack based on the raw measurement values, coupled with the electrochemical model, fluid dynamics model, and historical degradation data. The prediction results are quantified as a fuel cell stack health index, and the health index is sent to the centralized controller and cluster control unit at the control layer.

[0128] S3: The cluster control unit calculates real-time operating characteristic data based on the received raw measurement values, including voltage fluctuation rate, temperature gradient, and pressure change trend. It uses a lightweight AI inference module to make millisecond-level decisions based on the real-time operating characteristic data to generate single-stack emergency control instructions. It also generates a cluster load distribution plan based on the reinforcement learning algorithm, health index, and real-time operating characteristic data through a local dynamic load optimization model.

[0129] S4: The cluster control unit calculates the total cluster output power through a data synchronization mechanism, monitors for anomalies, and generates a fault flag. It only transmits the total cluster output power and the fault flag to the centralized controller. When the fault flag indicates that cross-cluster coordination is required, it triggers the generation of a global load distribution instruction.

[0130] S5: The centralized controller analyzes the fault flag to determine the location of the faulty cluster and the load shortfall, calculates the load compensation amount for the healthy cluster, which is equal to the load shortfall multiplied by the ratio of the healthy cluster's health index to the total health index, generates a global load allocation instruction, and sends it to the cluster control units of the healthy clusters, causing them to increase their hydrogen production load according to the compensation amount to maintain the stability of the system's total output power.

[0131] S6: The actuator group at the device layer executes the load distribution plan within the cluster, responds to the emergency control instructions of the single stack, and collects a new round of raw measurement values ​​to form a closed-loop control flow.

[0132] The digital twin-based coordinated control method for AEM hydrogen production stacks provided in the disclosed embodiments enables full lifecycle management of the AEM stack's operational status. The sensor and actuator groups at the device layer provide the system with high-precision real-time monitoring and rapid response capabilities, addressing the issue of insufficient dynamic adjustment in traditional control systems due to data acquisition delays or execution lags. The digital twin at the cloud platform layer integrates electrochemical models, fluid dynamics models, and a historical degradation data processing unit to construct a stack performance degradation prediction system. This allows the generation of a health index that combines the advantages of both physical mechanisms and data-driven approaches. These predictions provide a scientific basis for dynamic load optimization at the control layer, avoiding blind loading or overly conservative operational strategies. The cluster control unit at the control layer achieves multi-scale load distribution from a single stack to a cluster of stacks by collaborating with a local dynamic load optimization model and a global load distribution module. A reinforcement learning algorithm dynamically adjusts the load distribution strategy based on the health index and real-time operational characteristics, maximizing hydrogen production efficiency while ensuring stack safety. The data synchronization mechanism uses a lightweight protocol to transmit only critical information, significantly reducing communication overhead and shortening response time, while ensuring real-time awareness of the global status by the centralized controller. When a fault occurs, the linkage between the fault analysis module and the global load distribution module can quickly locate the fault source and complete the load compensation of the healthy cluster, effectively avoiding the impact of single point faults on system output.

[0133] For other technical details of the embodiments of the present disclosure, please refer to Figure 1 The embodiments shown are not described in detail here.

[0134] The present disclosure also discloses an electronic device, Figure 3 A structural block diagram of an electronic device according to an embodiment of the present disclosure is shown.

[0135] like Figure 3As shown, the electronic device includes a memory and a processor, wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor to implement the method according to an embodiment of the present disclosure.

[0136] The digital twin-based AEM hydrogen production stack collaborative control method includes the following steps:

[0137] S1: The sensor group at the equipment layer collects the raw measurement values ​​of voltage, temperature, and pressure of each AEM hydrogen production stack in real time, and synchronously uploads the raw measurement values ​​to the cloud platform layer and the cluster control unit at the control layer;

[0138] S2: The digital twin at the cloud platform layer predicts the performance degradation trend of each fuel cell stack based on the raw measurement values, coupled with the electrochemical model, fluid dynamics model, and historical degradation data. The prediction results are quantified as a fuel cell stack health index, and the health index is sent to the centralized controller and cluster control unit at the control layer.

[0139] S3: The cluster control unit calculates real-time operating characteristic data based on the received raw measurement values, including voltage fluctuation rate, temperature gradient, and pressure change trend. It uses a lightweight AI inference module to make millisecond-level decisions based on the real-time operating characteristic data to generate single-stack emergency control instructions. It also generates a cluster load distribution plan based on the reinforcement learning algorithm, health index, and real-time operating characteristic data through a local dynamic load optimization model.

[0140] S4: The cluster control unit calculates the total cluster output power through a data synchronization mechanism, monitors for anomalies, and generates a fault flag. It only transmits the total cluster output power and the fault flag to the centralized controller. When the fault flag indicates that cross-cluster coordination is required, it triggers the generation of a global load distribution instruction.

[0141] S5: The centralized controller analyzes the fault flag to determine the location of the faulty cluster and the load shortfall, calculates the load compensation amount for the healthy cluster, which is equal to the load shortfall multiplied by the ratio of the healthy cluster's health index to the total health index, generates a global load allocation instruction, and sends it to the cluster control units of the healthy clusters, causing them to increase their hydrogen production load according to the compensation amount to maintain the stability of the system's total output power.

[0142] S6: The actuator group at the device layer executes the load distribution plan within the cluster, responds to the emergency control instructions of the single stack, and collects a new round of raw measurement values ​​to form a closed-loop control flow.

[0143] Figure 4 A schematic diagram showing the structure of a computer system suitable for implementing the method according to an embodiment of the present disclosure is shown.

[0144] like Figure 4As shown, the computer system includes a processing unit, which can execute the various methods in the above-mentioned embodiments according to a program stored in a read-only memory (ROM) or a program loaded from a storage portion into a random access memory (RAM). In the RAM, various programs and data required for the operation of the computer system are also stored. The processing unit, ROM, and RAM are connected to each other via a bus. An input / output (I / O) interface is also connected to the bus.

[0145] The following components are connected to the I / O interface: an input part including a keyboard, a mouse, etc.; an output part including a cathode ray tube (CRT), a liquid crystal display (LCD), a speaker, etc.; a storage part including a hard disk, etc.; and a communication part including a network interface card such as a LAN card, a modem, etc. The communication part performs a communication process via a network such as the Internet. The drive is also connected to the I / O interface as needed. Removable media, such as magnetic disks, optical disks, magneto-optical disks, semiconductor memories, etc., are installed on the drive as needed so that the computer program read therefrom is installed into the storage part as needed. Among them, the processing unit can be implemented as a processing unit such as a CPU, a GPU, a TPU, an FPGA, an NPU, etc.

[0146] In particular, according to embodiments of the present disclosure, the methods described above can be implemented as computer software programs. For example, embodiments of the present disclosure include a computer program product comprising a computer program tangibly embodied on a machine-readable medium, the computer program comprising program code for executing the methods described above. In such embodiments, the computer program can be downloaded and installed from a network via a communication component and / or installed from a removable medium.

[0147] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, program segment or part of code, and the module, program segment or part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or can be implemented using a combination of dedicated hardware and computer instructions.

[0148] The units or modules involved in the embodiments described in this disclosure may be implemented by software or programmable hardware. The units or modules described may also be provided in a processor, and the names of these units or modules do not, in certain circumstances, constitute limitations on the units or modules themselves.

[0149] As another aspect, the present disclosure further provides a computer-readable storage medium. This computer-readable storage medium may be included in the electronic device or computer system described in the above embodiments, or may be a standalone computer-readable storage medium not incorporated into the device. The computer-readable storage medium stores one or more programs, which are used by one or more processors to execute the methods described in the present disclosure.

[0150] The above description is merely a preferred embodiment of the present disclosure and an illustration of the technical principles employed. Those skilled in the art should understand that the scope of the invention herein is not limited to the technical solutions formed by the specific combination of the above-mentioned technical features, but also encompasses other technical solutions formed by any combination of the above-mentioned technical features or their equivalents without departing from the inventive concept. For example, a technical solution formed by replacing the above-mentioned features with (but not limited to) technical features with similar functions disclosed in this disclosure.

Claims

1. A collaborative control system for AEM hydrogen production stack based on digital twin, characterized by: include: The equipment layer includes multiple fuel cell clusters, each of which contains multiple AEM hydrogen production fuel cells; Each AEM hydrogen production stack is equipped with a sensor group and an actuator group. The sensor group is used to collect the original measurement values ​​of the AEM hydrogen production stack in real time, and the actuator group is used to implement the load distribution plan within the cluster. The cloud platform layer includes a digital twin that integrates an electrochemical model, a fluid dynamics model, and a historical degradation data processing unit to receive the raw measurements and predict the performance degradation trend of each fuel cell stack, generating a prediction result quantified as a health index. The control layer includes a centralized controller and multiple cluster control units; Each cluster control unit includes a real-time operation feature calculation module, a local dynamic load optimization model, and a data synchronization mechanism; the real-time operation feature calculation module is used to calculate the real-time operation feature data based on the received raw measurement values; the local dynamic load optimization model generates the load distribution plan within the cluster based on the reinforcement learning algorithm, the health index, and the real-time operation feature data; the data synchronization mechanism is used to calculate the total cluster output power and monitor anomalies to generate fault flags, and only transmit the total cluster output power and fault flags to the centralized controller; The centralized controller includes a fault analysis module and a global load distribution module; the fault analysis module is used to analyze the fault sign to determine the location of the fault cluster and the load gap; the global load distribution module is used to calculate the load compensation amount of the healthy cluster, generate a global load distribution instruction and send it to the cluster control unit of the healthy cluster, so that the actuator group of the healthy cluster executes the cluster load distribution plan to maintain the stability of the total output power of the cluster.

2. The AEM hydrogen production stack collaborative control system based on edge intelligence according to claim 1 is characterized in that: The health index H generated by the digital twin i Calculated by the following formula: Where, ΔV i is the voltage offset, V nom Indicates rated voltage, T i is the real-time temperature, and top is the cumulative running hours.

3. The AEM hydrogen production stack collaborative control system based on edge intelligence according to claim 1 is characterized in that: The data packet transmitted by the data synchronization mechanism consists of 5 bytes, wherein the first 4 bytes represent the single-precision floating point number of the total cluster output power, and the fifth byte represents the fault flag, which includes normal state, warning state and serious fault state.

4. The AEM hydrogen production stack collaborative control system based on edge intelligence according to claim 1 is characterized in that: Also includes: The dynamic load distribution module is used to generate load distribution strategies based on the health index generated by the digital twin.

5. The AEM hydrogen production stack collaborative control system based on edge intelligence according to claim 4 is characterized in that: The load distribution strategy is as follows: When the health index H i When it is greater than or equal to 0.8, 95% to 100% of the baseline load is assigned; When the health index H i When it is greater than or equal to 0.6 and less than 0.8, 70 to 95 percent of the baseline load is assigned; When the health index H i When it is less than 0.6, no more than 60 percent of the baseline load is distributed.

6. The AEM hydrogen production stack collaborative control system based on edge intelligence according to claim 1 is characterized in that: The part of the global load distribution module that calculates the load compensation amount of the healthy cluster, generates the global load distribution instruction, and sends it to the cluster control unit of the healthy cluster is configured as follows: Eliminate faulty clusters and identify all healthy clusters; For each health cluster, the amount of load compensation that should be increased is calculated according to the proportion of its health index in the total health index; The calculated load compensation amount is distributed to the cluster control units of the corresponding healthy clusters.

7. The AEM hydrogen production stack collaborative control system based on edge intelligence according to claim 1 is characterized in that: Each cluster control unit also includes a lightweight AI reasoning module for making millisecond-level decisions based on the real-time operating characteristic data to generate single-stack emergency control instructions.

8. The AEM hydrogen production stack collaborative control system based on edge intelligence according to claim 7 is characterized in that: The lightweight AI inference module deploys and runs a quantized deep Q network model; The model size of the quantized deep Q network model is no more than 8 megabytes, and the inference delay is no more than 8 milliseconds. The reward function of the quantized deep Q network model is a times the hydrogen production efficiency minus b times the voltage fluctuation minus c times the number of overtemperatures; where the sum of the values ​​of a, b, and c is 1.

9. A collaborative control method for an AEM hydrogen production stack based on digital twins, characterized in that: The following steps are involved: S1: The sensor group at the equipment layer collects the raw measurement values ​​of voltage, temperature, and pressure of each AEM hydrogen production stack in real time, and synchronously uploads the raw measurement values ​​to the cloud platform layer and the cluster control unit at the control layer; S2: The digital twin at the cloud platform layer predicts the performance degradation trend of each fuel cell stack based on the raw measurement values, coupled with the electrochemical model, fluid dynamics model, and historical degradation data. The prediction results are quantified as a fuel cell stack health index, and the health index is sent to the centralized controller and cluster control unit at the control layer. S3: The cluster control unit calculates real-time operating characteristic data based on the received raw measurement values, including voltage fluctuation rate, temperature gradient, and pressure change trend. It uses a lightweight AI inference module to make millisecond-level decisions based on the real-time operating characteristic data to generate single-stack emergency control instructions. It also generates a cluster load distribution plan based on the reinforcement learning algorithm, health index, and real-time operating characteristic data through a local dynamic load optimization model. S4: The cluster control unit calculates the total cluster output power through the data synchronization mechanism, monitors anomalies and generates a fault flag, and only transmits the total cluster output power and the fault flag to the centralized controller; When the fault flag indicates that cross-cluster coordination is required, the generation of global load distribution instructions is triggered; S5: The centralized controller analyzes the fault flag to determine the location of the faulty cluster and the load shortfall, calculates the load compensation amount for the healthy cluster, which is equal to the load shortfall multiplied by the ratio of the healthy cluster's health index to the total health index, generates a global load allocation instruction, and sends it to the cluster control units of the healthy clusters, causing them to increase their hydrogen production load according to the compensation amount to maintain the stability of the system's total output power. S6: The actuator group at the device layer executes the load distribution plan within the cluster, responds to the emergency control instructions of the single stack, and collects a new round of raw measurement values ​​to form a closed-loop control flow.

Images