Fuel cell stack temperature-flow rate coordinated real-time optimization control method and system

Abstract

The application aims to provide a fuel cell stack temperature-flow collaborative real-time optimization control method and system, and relates to the technical field of fuel cells. By constructing a high-speed synchronous data link, microsecond-level synchronous acquisition and low-delay transmission of entity stack operation state data are realized; by presetting a high-precision dynamic digital twin model and dynamically correcting key parameters online, the model can track the state evolution and characteristic drift of the entity stack in real time, and always maintain high-precision synchronization with the physical entity; based on the high-credibility model, a model predictive control framework is used to solve multi-objective optimization problems, and a collaborative control instruction sequence of hydrogen flow, air flow and cooling liquid flow is directly generated, realizing end-to-end closed-loop optimization from data acquisition to control decision, and fundamentally solving the problem that temperature and flow control are mutually fragmented and difficult to collaborate in traditional methods, and the stack temperature uniformity and system efficiency can be effectively improved under dynamic working conditions.

Description

Technical Field

[0001] This invention relates to the field of fuel cell technology, and in particular to a method and system for real-time optimization control of temperature and flow rate in fuel cell stacks. Background Technology

[0002] Proton exchange membrane fuel cell stacks have broad application prospects in transportation, stationary power generation, and other fields. During dynamic operation, rapid changes in load power lead to drastic fluctuations in the heat generation rate and reactant demand of the electrochemical reaction within the stack, while temperature and flow rate are strongly coupled and time-delayed. Existing control methods, such as traditional PID split-loop control, struggle to handle multivariable strongly coupled systems, easily exhibiting control lag and overshoot. Model predictive control research largely focuses on single-objective control, failing to achieve true synergistic optimization of temperature and flow rate. In digital twin-based system management schemes, the digital twin model primarily serves as a state monitoring tool rather than a core engine directly driving real-time synergistic optimization. Therefore, existing technologies struggle to achieve precise matching between uniform temperature field distribution and reactant flow rate under dynamic operating conditions, hindering further improvements in fuel cell system performance and reliability. Summary of the Invention

[0003] The purpose of this invention is to provide a real-time optimization control method and system for temperature-flow coordination in fuel cell stacks. By constructing a high-speed synchronous data link, microsecond-level synchronous acquisition and low-latency transmission of physical stack operating status data are achieved, providing a precise spatiotemporal data foundation for subsequent modeling and control. By pre-setting a high-precision dynamic digital twin model and dynamically correcting key parameters online, the model can track the state evolution and characteristic drift of the physical stack in real time, always maintaining high-precision synchronization with the physical entity, thereby providing reliable predictive capabilities for control decisions. Based on this high-reliability model, a model predictive control framework is used to solve multi-objective optimization problems in a rolling manner, directly generating a coordinated control command sequence for hydrogen, air, and coolant flow rates. This achieves end-to-end closed-loop optimization from data acquisition to control decision-making, fundamentally solving the problem of the disconnect and difficulty in coordination between temperature and flow control in traditional methods. Under dynamic operating conditions, it can effectively improve the stack temperature uniformity and system efficiency.

[0004] In a first aspect, the present invention provides a real-time optimization control method for temperature-flow coordinated operation of a fuel cell stack, comprising:

[0005] Real-time operating status data of the physical fuel cell stack is collected synchronously via a high-speed synchronous data link.

[0006] Real-time operating status data is input into a preset high-precision dynamic digital twin model, and the model is run to predict the future internal state of the physical fuel cell stack in real time. During the operation, the key parameters of the model are dynamically corrected online based on the deviation between the real-time operating status data and the model prediction values, so that the model always maintains high-precision synchronization with the physical fuel cell stack.

[0007] Based on the model-predicted future internal state, a model predictive control framework is adopted to solve a multi-objective optimization problem in each control cycle, generating a coordinated control command sequence for hydrogen flow, air flow, and coolant flow. The multi-objective optimization problem aims at least to minimize the internal temperature non-uniformity of the fuel cell stack and optimize system efficiency.

[0008] The first instruction in the coordinated control instruction sequence obtained from the current control cycle is sent to the actuator via a high-speed synchronous data link to adjust the reactant flow rate of the physical fuel cell stack in real time.

[0009] In some preferred embodiments of the present invention, the step of synchronously acquiring real-time operating status data of a physical fuel cell stack via a high-speed synchronous data link includes:

[0010] Distributed synchronous acquisition nodes are deployed in multiple key physical regions of the physical fuel cell stack; among them, the key physical regions include at least one of the following: cathode flow channel outlet, coolant inlet and outlet, air intake manifold, and multiple representative cross sections along the length of the stack.

[0011] A global synchronization pulse signal is generated by the master clock source. The global synchronization pulse signal triggers all acquisition nodes to start sampling on the same hardware clock edge, so as to synchronously acquire temperature, pressure, flow and voltage data of key physical areas with microsecond-level precision.

[0012] Each acquisition node preprocesses the acquired raw signal and adds a precise timestamp to obtain a time-stamped data packet;

[0013] Time-stamped data packets are transmitted to the data service layer via a network transport layer based on deterministic real-time Ethernet.

[0014] In some preferred embodiments of the present invention, the high-precision dynamic digital twin model includes: a multiphysics coupling mechanism kernel; the multiphysics coupling mechanism kernel includes at least an electrochemical sub-model, a computational fluid dynamics sub-model, and an unsteady heat transfer sub-model, and the electrochemical sub-model, the computational fluid dynamics sub-model, and the unsteady heat transfer sub-model are co-calculated using a bidirectional strong coupling method; after the step of inputting real-time operating status data into the preset high-precision dynamic digital twin model, the method further includes:

[0015] The computational fluid dynamics sub-model calculates the gas velocity, gas pressure, and gas component concentration distribution within the flow channel based on the current state.

[0016] Gas velocity, gas pressure, and gas component concentration distribution are input into the electrochemical sub-model, which calculates the spatially distributed current density and local reaction heat generation rate.

[0017] The local reaction heat generation rate is injected as an internal heat source into the unsteady heat transfer sub-model to drive the unsteady heat transfer sub-model to calculate the three-dimensional transient temperature field inside the stack.

[0018] The calculated three-dimensional transient temperature field is fed back to the electrochemical sub-model and the computational fluid dynamics sub-model to update the temperature-dependent electrochemical dynamic parameters in the electrochemical sub-model and the temperature-dependent fluid property parameters in the computational fluid dynamics sub-model. The calculation is iterated until convergence.

[0019] In some preferred embodiments of the present invention, the step of dynamically correcting key parameters of the model online based on the deviation between real-time running status data and model predictions during operation includes:

[0020] Within each control cycle, a recursive closed-loop process of prediction-comparison-update is executed, including:

[0021] Based on the updated key parameter estimates from the previous time step and the input at the current time step, the multiphysics coupling mechanism kernel is driven to perform a forward simulation to calculate the observation predictions corresponding to the current time step.

[0022] The predicted observation value is compared with the real-time sensor observation value obtained through the high-speed synchronous data link at the current moment, and the prediction error vector between the predicted observation value and the real-time sensor observation value is calculated.

[0023] A recursive least squares algorithm with a forgetting factor is used to calculate the gain matrix in real time based on the prediction error vector and the pre-calculated sensitivity matrix, and to update the estimated values ​​of key parameters based on the gain matrix; among which, the key parameters include at least: cathode exchange current density and membrane conductivity parameters.

[0024] The updated key parameter estimates are injected into the multiphysics coupling mechanism kernel in real time to replace the old values, enabling the model to run in the corrected state in the next control cycle.

[0025] In some preferred embodiments of the present invention, the step of using a model predictive control framework to solve a multi-objective optimization problem on a rolling basis within each control cycle, based on model-predicted future internal states, includes:

[0026] At the beginning of each control cycle, the current physical stack state data acquired through a high-speed synchronous data link is used as the initial state of the high-precision dynamic digital twin model to complete the state synchronization between the virtual model and the physical entity.

[0027] Set a prediction time domain that is greater than or equal to the control time domain, and use a high-precision dynamic digital twin model after state synchronization to predict the future state response of the system corresponding to different candidate control sequences applied in the prediction time domain.

[0028] An online solution is provided for a multi-objective optimization problem within a finite time domain to find the optimal control sequence that minimizes the objective function value within the control time domain. The objective function includes at least a first weight term penalizing the deviation between the maximum temperature difference of the fuel cell cross-section and a preset target value, a second weight term penalizing the deviation between the system net power and a preset reference value, and a third smoothness weight term penalizing the amplitude of changes in control actions. The decision variables of the optimization problem are the setpoints for hydrogen flow rate, air flow rate, and coolant flow rate at each step within the future control time domain. The solution process is constrained by preset constraints, which include at least system dynamic constraints described by a high-precision dynamic digital twin model, safety range constraints for the flow rate and rotational speed of each actuator, constraints on the rate of change of control commands, and safety boundary constraints for key state variables.

[0029] In some preferred embodiments of the present invention, the maximum temperature difference of the fuel cell stack cross section is determined by post-processing the three-dimensional temperature field data output from the high-precision dynamic digital twin model and calculating the difference between the maximum and minimum temperatures on multiple key cross sections inside the fuel cell stack; the net power of the system is determined by the difference between the output power of the fuel cell stack and the power consumed by all auxiliary systems.

[0030] In some preferred embodiments of the present invention, a multi-objective optimization problem in a finite time domain is solved online using a real-time iterative sequential quadratic programming algorithm. A preset number of iterations are performed in each control cycle to ensure that the solution is completed and the optimal control sequence is output within the preset control cycle time.

[0031] In some preferred embodiments of the present invention, the method further includes:

[0032] After executing the optimal control command for the current moment, at the start of the next control cycle, the predicted trajectory of the previous cycle is no longer used. Instead, the physical stack state data of the current moment, obtained through actual measurement via a high-speed synchronous data link, is used as the initial state for a new round of rolling optimization.

[0033] In some preferred embodiments of the present invention, the method further includes:

[0034] Continuously monitor the prediction accuracy of the high-precision dynamic digital twin model during recent operation. When the average value of the prediction accuracy index in the sliding window exceeds the preset warning threshold, the background offline learning process is automatically triggered.

[0035] In the background offline learning process, a representative segment of running data is automatically extracted from the historical database. With the goal of minimizing the overall prediction error of the model on the representative running data, an intelligent optimization algorithm is used to further optimize the hyperparameters of the online parameter dynamic correction module and some fixed parameters in the kernel of the multiphysics coupling mechanism.

[0036] The validated and optimized parameter set will be safely injected into and replace the original parameters in the high-precision dynamic digital twin model via hot update without interrupting the normal operation of the system.

[0037] In a second aspect, the present invention provides a real-time optimization control system for temperature-flow coordination in a fuel cell stack, comprising:

[0038] The data acquisition module is used to synchronously acquire real-time operating status data of the physical fuel cell stack via a high-speed synchronous data link;

[0039] The model prediction module is used to input real-time operating status data into a preset high-precision dynamic digital twin model and run the model to predict the future internal state of the physical fuel cell stack in real time. During the operation, based on the deviation between the real-time operating status data and the model prediction value, the key parameters of the model are dynamically corrected online to ensure that the model always maintains high-precision synchronization with the physical fuel cell stack.

[0040] An optimization control module is used to predict the future internal state based on the model. It adopts a model predictive control framework to solve a multi-objective optimization problem in each control cycle and generate a coordinated control command sequence for hydrogen flow, air flow, and coolant flow. The multi-objective optimization problem has at least the optimization objectives of minimizing the internal temperature non-uniformity of the fuel cell stack and optimizing the system efficiency.

[0041] The instruction execution module is used to send the first instruction in the cooperative control instruction sequence obtained from the current control cycle to the actuator via a high-speed synchronous data link, so as to adjust the reactant flow rate of the physical fuel cell stack in real time.

[0042] This invention brings the following beneficial effects:

[0043] The purpose of this invention is to provide a real-time optimization control method and system for temperature-flow coordination in fuel cell stacks. By constructing a high-speed synchronous data link, microsecond-level synchronous acquisition and low-latency transmission of physical stack operating status data are achieved, providing a precise spatiotemporal data foundation for subsequent modeling and control. By pre-setting a high-precision dynamic digital twin model and dynamically correcting key parameters online, the model can track the state evolution and characteristic drift of the physical stack in real time, always maintaining high-precision synchronization with the physical entity, thereby providing reliable predictive capabilities for control decisions. Based on this high-reliability model, a model predictive control framework is used to solve multi-objective optimization problems in a rolling manner, directly generating a coordinated control command sequence for hydrogen, air, and coolant flow rates. This achieves end-to-end closed-loop optimization from data acquisition to control decision-making, fundamentally solving the problem of the disconnect and difficulty in coordination between temperature and flow control in traditional methods. Under dynamic operating conditions, it can effectively improve the stack temperature uniformity and system efficiency. Attached Figure Description

[0044] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0045] Figure 1 A flowchart of a real-time optimization control method for temperature-flow coordinated operation of a fuel cell stack, provided in an embodiment of the present invention;

[0046] Figure 2 This is a schematic diagram of a high-speed synchronous data link workflow provided in an embodiment of the present invention;

[0047] Figure 3 A flowchart illustrating the overall process of model construction and operation provided in this embodiment of the invention;

[0048] Figure 4 A schematic diagram of the workflow of a multivariable collaborative real-time optimization controller provided in an embodiment of the present invention;

[0049] Figure 5 This is a schematic diagram of a real-time optimization control system for temperature-flow coordination in a fuel cell stack, provided in an embodiment of the present invention.

[0050] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0051] Icons: 310 - Data acquisition module; 320 - Model prediction module; 330 - Optimization control module; 340 - Instruction execution module; 400 - Memory; 401 - Processor; 402 - Bus; 403 - Communication interface. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0053] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0054] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0055] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product of this invention is in use. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention. In addition, the terms "first," "second," "third," etc., are only used to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0056] Furthermore, terms such as "horizontal," "vertical," and "sag" do not imply that components must be absolutely horizontal or suspended, but rather that they can be slightly tilted. For example, "horizontal" simply means that its direction is more horizontal relative to "vertical," and does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0057] In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0058] The following detailed description of some embodiments of the present invention is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0059] Example 1

[0060] This invention provides a real-time optimization control method for temperature-flow coordinated operation of fuel cell stacks, see [link to relevant documentation]. Figure 1 The flowchart shown in this embodiment of the invention provides a real-time optimization control method for temperature-flow coordinated operation of a fuel cell stack. The method includes:

[0061] Step S102: Real-time operating status data of the physical fuel cell stack is collected synchronously via a high-speed synchronous data link.

[0062] Specifically, through distributed synchronous acquisition nodes deployed in key physical areas of the physical fuel cell stack, and based on hardware-level synchronous triggering and deterministic real-time Ethernet technology, microsecond-level synchronous acquisition and low-latency transmission of multi-source sensor data are achieved, providing a precise spatiotemporal data foundation for subsequent digital twin modeling and optimized control. See also Figure 2 The diagram shown is a schematic of a high-speed synchronous data link workflow provided by an embodiment of the present invention. The overall architecture is divided into a physical sensing layer, a network transmission layer, and a data service layer.

[0063] Furthermore, in some preferred embodiments of the present invention, the step of synchronously acquiring real-time operating status data of a physical fuel cell stack via a high-speed synchronous data link includes: deploying distributed synchronous acquisition nodes in multiple key physical areas of the physical fuel cell stack; wherein, the key physical areas include at least one of the following: cathode flow channel outlet, coolant inlet and outlet, air intake manifold, and multiple representative cross-sections along the length of the stack; generating a global synchronization pulse signal from a master clock source, triggering all acquisition nodes to start sampling at the same hardware clock edge through the global synchronization pulse signal, and synchronously acquiring temperature, pressure, flow rate, and voltage data of the key physical areas with microsecond-level precision; each acquisition node preprocessing the acquired raw signal and adding a precise timestamp to obtain a time-stamped data packet; and transmitting the time-stamped data packet to the data service layer through a network transport layer based on deterministic real-time Ethernet.

[0064] For details, please refer to [link / reference]. Figure 2 Synchronous acquisition nodes: Intelligent data acquisition nodes are deployed in multiple key areas of the fuel cell stack (such as the intake manifold, coolant circuit, and mid-section of the stack). Each node integrates a multi-channel ADC (analog-to-digital converter) and can synchronously acquire analog signals such as temperature, pressure, and voltage in its local area.

[0065] Hardware synchronization triggering: All acquisition nodes are connected via a single hardware trigger line of equal length (or a hardware clock output based on IEEE 1588 PTP). A global synchronization pulse signal is emitted by the master clock source, and all nodes start sampling on the same hardware clock edge upon receiving the pulse, fundamentally eliminating the millisecond-level time difference caused by traditional polling acquisition methods.

[0066] The nodes have basic data processing capabilities, such as real-time digital filtering based on SG filters, which can achieve on-site preprocessing, remove high-frequency noise before transmission, and reduce network load.

[0067] The temperature sensor uses multiple platinum resistance thermometers, with multiple sensors arranged at the coolant inlet, outlet, and cathode flow channel outlets at multiple representative cross-sections along the stack length. Sensors are also arranged at the anode inlet and return gas mains, and at key cooling pipeline nodes. Mass flow meters and pressure sensors are installed in the air inlet main, hydrogen inlet main, and coolant main circuit, respectively. The voltage monitoring module can simultaneously collect the voltage of all individual cells, and the sampling frequency meets the real-time control requirements.

[0068] A global synchronization pulse signal is generated by the master clock source. This global synchronization pulse signal triggers all acquisition nodes to start sampling on the same hardware clock edge, synchronously acquiring temperature, pressure, flow, and voltage data of the key physical area with microsecond-level precision. Each acquisition node preprocesses the acquired raw signal and adds a precise timestamp to obtain a time-stamped data packet. The preprocessing includes real-time digital filtering using an SG filter to remove high-frequency noise.

[0069] The time-stamped data packets are transmitted to the data service layer via a network transport layer based on deterministic real-time Ethernet. The deterministic real-time Ethernet can be implemented using the EtherCAT protocol as an alternative, or employ time-sensitive networking, reserving a fixed time window for the real-time data stream through a time-aware shaper. After receiving the data packets, the data service layer aligns and packages the data from the same sampling moment according to a unified time reference, forming structured "system snapshot" data frames, which are then sent to the subsequent model and controller via a reliable transmission protocol.

[0070] See also Figure 2The preferred communication protocol is EtherCAT, which adopts a "fly-read-fly-write" communication mechanism. When the data frame sent by the master station passes through each slave station (acquisition node, controller, driver), the slave station reads or writes the specified data in real time. One frame traverses all devices, and the typical cycle time can be as low as 100μs. The synchronization jitter is less than 1μs, which perfectly meets the microsecond-level synchronization requirements.

[0071] The alternative, TSN, is based on the IEEE 802.1 Time-Sensitive Networking series of standards for standard Ethernet. It uses a Time-Aware Shaper (TAS) to reserve a fixed time window for real-time data streams, ensuring that their transmission is not affected by other background traffic. It also supports the standard TCP / IP protocol stack, facilitating integration with upper-layer information systems.

[0072] The network topology uses either a star or linear topology. The master station (industrial PC or high-end PLC) acts as the central controller, connecting all slave devices via switches. The network is configured with redundant paths to ensure that a single point of failure does not affect critical data transmission.

[0073] Clock synchronization is achieved by implementing the IEEE 1588 Precision Time Protocol (PTP) across the entire network. The network switch acts as a transparent clock or boundary clock, correcting the dwell time of PTP messages and accurately transmitting the master clock (usually GPS or a high-stability crystal oscillator) to every sensor node and actuator driver, thus achieving sub-microsecond synchronization of clocks across the entire network.

[0074] By using hardware-level synchronous triggering and deterministic network transmission, the sensor data distributed throughout the entire stack is ensured to have a unified time reference at the microsecond level. This fundamentally eliminates the spatiotemporal misalignment problem caused by asynchronous data acquisition, providing a high-quality data foundation for building accurate spatiotemporal state mapping for the digital twin model and ensuring the reliability of all subsequent analysis and control.

[0075] See also Figure 2 The service module performs timestamp alignment and data frame assembly, receiving raw data with PTP timestamps from each acquisition node. Based on a unified time base, the service module aligns and packages data from different physical locations at the same sampling time (e.g., the kth millisecond of absolute time T) to form a structured "system snapshot" data frame. This frame contains timestamps, all sensor IDs, and their values.

[0076] For transmission and verification, lightweight but reliable transport protocols (such as a custom reliable transport protocol based on UDP or RTPS) are used to send data frames to the digital twin model building unit and MPC controller. Each data packet contains a cyclic redundancy check code to ensure data integrity. A "heartbeat" and "acknowledgment" mechanism is established to monitor link connectivity in real time.

[0077] The control command encoding and distribution process involves receiving optimized commands (such as flow setpoints) from the MPC controller, encoding them into command frames that conform to the actuator communication protocol (such as EtherCAT CoE), and distributing them to the corresponding frequency converters and proportional valve drives through the same deterministic network. Simultaneously, the actuators are required to provide feedback on actual executed values ​​(such as valve opening degree) to form a closed-loop verification.

[0078] Step S104: Input the real-time operating status data into the preset high-precision dynamic digital twin model and run the model to predict the future internal state of the physical fuel cell stack in real time; wherein, during the operation, based on the deviation between the real-time operating status data and the model prediction value, the key parameters of the model are dynamically corrected online to ensure that the model always maintains high-precision synchronization with the physical fuel cell stack.

[0079] Specifically, by using a pre-constructed multi-physics coupling mechanism kernel combined with an online parameter dynamic correction module based on recursive least squares, the digital twin model can track the state evolution and characteristic drift of the physical fuel cell stack in real time, providing a high-fidelity predictive basis for subsequent optimized control. See also Figure 3 The embodiment of the present invention shown provides a general flowchart for model construction and operation, which includes two stages: offline pre-training and online synchronization.

[0080] Furthermore, in some preferred embodiments of the present invention, the high-precision dynamic digital twin model includes: a multiphysics coupling mechanism kernel; the multiphysics coupling mechanism kernel includes at least an electrochemical sub-model, a computational fluid dynamics sub-model, and an unsteady heat transfer sub-model, and the electrochemical sub-model, the computational fluid dynamics sub-model, and the unsteady heat transfer sub-model are co-calculated using a bidirectional strong coupling method; after the step of inputting real-time operating state data into the preset high-precision dynamic digital twin model, the method further includes: the computational fluid dynamics sub-model calculating the gas velocity, gas pressure, and gas flow rate within the flow channel based on the current state. The gas component concentration distribution is calculated. Gas velocity, gas pressure, and gas component concentration distribution are input into the electrochemical sub-model, which calculates the spatially distributed current density and local reaction heat generation rate. The local reaction heat generation rate is injected as an internal heat source into the unsteady-state heat transfer sub-model, driving it to calculate the three-dimensional transient temperature field inside the stack. The calculated three-dimensional transient temperature field is fed back to the electrochemical sub-model and the computational fluid dynamics sub-model to update the temperature-dependent electrochemical dynamic parameters in the electrochemical sub-model and the temperature-dependent fluid property parameters in the computational fluid dynamics sub-model. The calculation is iterative until convergence.

[0081] Specifically, the electrochemical sub-model employs a discretization method combining one-dimensional discretization along the flow channel direction and quasi-two-dimensional discretization perpendicular to the membrane electrode direction. Its core equations include the expression for the battery output voltage:

[0082] ;

[0083] in, Thermodynamic reversible potential related to temperature and partial pressure of reactant gases; The activation overpotential is described by the Butler-Volmer equation, and its key parameter is the exchange current density. ; To be related to membrane resistance The relevant ohmic overpotential; To characterize the concentration overpotential limiting reactant transport, this model is used to solve nonlinear equations under given local temperature, gas concentration, and potential conditions. Iteration yields local value.

[0084] Based on the traditional output voltage formula, a discretization method using one-dimensional flow path and quasi-two-dimensional perpendicular to the membrane electrode direction is employed to calculate the current density distribution and local reaction heat generation rate. This significantly improves computational efficiency while maintaining accuracy, enabling high-fidelity electrochemical models to complete simulations within seconds, meeting real-time control requirements. It can more realistically reflect the spatial non-uniformity of current density and reaction heat distribution, providing more accurate input for temperature field prediction.

[0085] Discretization is performed along the flow channel direction (one-dimensional) and perpendicular to the membrane electrode direction (quasi-two-dimensional, considering the gas diffusion layer), based on the current density. Distribution, obtaining the local reaction heat generation rate:

[0086] ;

[0087] The computational fluid dynamics and proton transfer model is responsible for simulating gas flow, pressure distribution, component transport, and water management processes within a bipolar channel. To balance accuracy and velocity, this model employs a combination of a simplified equivalent flow resistance network for a three-dimensional channel and a one-dimensional convection-diffusion equation. Its mass conservation equation can be expressed as:

[0088] ;

[0089] in, This indicates the density of a gas mixture (such as air, hydrogen, water vapor, etc.) at a given location and time. This refers to the mass fraction of the component. For fluid velocity vector; The effective diffusion coefficient of component k; The source term for component k consumed or generated in an electrochemical reaction.

[0090] This method simplifies the traditional 3D CFD model by combining an equivalent flow resistance network of a 3D flow channel with a one-dimensional convection-diffusion equation. This significantly reduces computational complexity and avoids the long computation time associated with traditional 3D CFD models. Key physical processes (such as pressure drop and component distribution) are preserved, enabling real-time calculations while maintaining prediction accuracy.

[0091] Momentum transfer is efficiently calculated using pressure drop formulas based on Darcy's law or empirical correlations:

[0092] ;

[0093] in, Pressure drop (pressure loss); It is the coefficient of friction, and it is the Reynolds number. The function; is the pipe length; L is the hydraulic diameter; The fluid density is given.

[0094] The unsteady heat transfer sub-model serves as the core for achieving accurate temperature field prediction. It calculates the temperature distribution of solid components and fluids within the fuel cell stack by solving the three-dimensional energy equation.

[0095] The energy equation for the solid domain is:

[0096] ;

[0097] The energy equation for the fluid domain is:

[0098] ;

[0099] At the gas-solid and liquid-solid interfaces, strong coupled heat transfer is achieved through dynamically calculated convective heat transfer coefficient h, with the following boundary conditions:

[0100] ;

[0101] in, The density of the solid; The density of the fluid; is the specific heat capacity at constant pressure of a solid; The specific heat capacity at constant pressure of the fluid; The temperature field of the solid; For the temperature field of the fluid; k s k is the thermal conductivity of the solid. f The thermal conductivity of the fluid; denoted as the solid internal heat source (heat generation rate per unit volume); u is the fluid velocity vector; n is the interface unit normal vector (usually pointing from the solid to the fluid); and h is the interface convective heat transfer coefficient.

[0102] The three sub-models mentioned above are coupled together to form an organic whole: Under a given operating condition, the CFD sub-model first calculates the gas velocity, pressure, and component concentration distribution within the flow channel; these distribution results are then passed as input to the electrochemical sub-model, which calculates the spatially distributed current density. With the rate of heat generation of reaction These heat sources are injected into the heat transfer model, thereby driving the calculation of the three-dimensional temperature field; the updated temperature field, in turn, affects the kinetic parameters of the electrochemical reaction (such as exchange current density). ) and the transport properties of the membrane (such as membrane resistance) This also changes the fluid's physical properties, forming a tight closed loop that is solved iteratively at each time step until convergence.

[0103] By constructing a two-way strongly coupled computational mechanism of electrochemistry, fluid dynamics and heat transfer, the model can realistically reflect the dynamic interaction between various physical fields inside the fuel cell stack, solving the problem that simplified models cannot accurately characterize complex coupling effects. At the same time, through reasonable model dimensionality reduction design, the computational speed requirements for real-time control are met while ensuring prediction accuracy, achieving a balance between high fidelity and real-time performance.

[0104] Furthermore, in some preferred embodiments of the present invention, the step of dynamically correcting the key parameters of the model online based on the deviation between the real-time operating status data and the model prediction values ​​during operation includes: executing a recursive closed-loop process of prediction-comparison-update in each control cycle, including: driving the multiphysics coupling mechanism kernel to perform a forward simulation based on the updated key parameter estimates from the previous moment and the input at the current moment, calculating the observed prediction values ​​corresponding to the current moment; comparing the observed prediction values ​​with the real-time sensor observation values ​​obtained through a high-speed synchronous data link at the current moment, calculating the prediction error vector between the observed prediction values ​​and the real-time sensor observation values; using a recursive least squares algorithm with a forgetting factor, calculating the gain matrix in real time based on the prediction error vector and the pre-calculated sensitivity matrix, and updating the estimated values ​​of the key parameters based on the gain matrix; wherein, the key parameters include at least: cathode exchange current density and membrane conductivity parameters; and injecting the updated key parameter estimates into the multiphysics coupling mechanism kernel in real time to replace the old values, so that the model can operate in the corrected state in the next control cycle.

[0105] Specifically, for key parameters in the mechanistic model that are difficult to measure accurately or drift over time (such as exchange current density and membrane conductivity), the recursive least squares (RLS) algorithm is used for calculation. This module uses real-time sensor data (such as total voltage and outlet temperature) as observations, and continuously and recursively back-calculates and updates these internal parameters, thereby dynamically compensating for model errors and system aging drift, ensuring that the mechanistic model always maintains a high-precision mapping with the physical object.

[0106] Observable quantities: quantities that are easy to measure directly and are sensitive to correction parameters, such as the total voltage V of the fuel cell stack. stack Outlet coolant temperature T cool-out Temperature T at a specific location sensor .

[0107] This module executes synchronously with the main control loop within each control cycle (e.g., 1 second), forming a recursive "prediction-comparison-update" closed loop. The detailed steps are as follows:

[0108] Step 1: Data Synchronization and Preparation: At time k, the module receives real-time sensor observations y from the high-speed data link. m (k) (such as total stack voltage, key point temperature), and simultaneously acquire the current state and input (such as current, flow rate setpoint) of the digital twin mechanism model.

[0109] Step 2: Forward prediction using the model: Utilize the parameter estimates updated in the previous time step based on the mechanistic model. Given the current input, run a forward simulation once to calculate the corresponding observed predictions. .

[0110] Step 3: Prediction Error Calculation: Calculate the error vector between the predicted and measured values.

[0111] ;

[0112] Step 4: Recursive parameter update, calculate the gain matrix:

[0113] ;

[0114] in, This is the sensitivity matrix (Jacobi matrix), which represents the sensitivity of the observed values ​​to changes in the parameter. λ is the parameter estimation error covariance matrix; λ is the forgetting factor, used to give new data higher weight, so that the algorithm can track time-varying parameters.

[0115] Update parameter estimates:

[0116] ；

[0117] Update the covariance matrix:

[0118] ;

[0119] Step 5: Hot reset of model parameters: Reset the updated parameters Immediate and seamless injection of mechanistic models (e.g., updating the exchange current density i0 and membrane conductivity parameters in the electrochemical sub-model) allows the model to run immediately in the corrected state in the next cycle.

[0120] Among them, the key parameters include at least: cathode exchange current density i 0cathode and membrane conductivity σ mem The forgetting factor λ is typically set to 0.95~0.999, and in this embodiment, it is set to 0.995. The initial parameter covariance matrix P(0) is set to a large diagonal matrix, for example, 10. 6 ×I indicates that the initial uncertainty is large.

[0121] Iterative Loop: Upon reaching time k+1, repeat steps 1 to 5. This process runs continuously online, dynamically correcting key parameters online using a recursive least squares algorithm. This enables the digital twin model to track the characteristic drift of the fuel cell stack caused by aging, decay, or environmental changes in real time, solving the problem of accuracy degradation in traditional fixed-parameter models over long-term operation and ensuring that the model prediction error remains stable within 10% for a long period.

[0122] Step S106: Based on the model-predicted future internal state, a model predictive control framework is adopted to solve the multi-objective optimization problem in each control cycle, generating a coordinated control command sequence for hydrogen flow rate, air flow rate, and coolant flow rate; wherein, the multi-objective optimization problem has at least the optimization objectives of minimizing the internal temperature non-uniformity of the fuel cell stack and optimizing the system efficiency.

[0123] Specifically, within each control cycle, the digital twin model is initialized with the current measured state. Utilizing its predictive capabilities, a multi-objective optimization problem within a finite time domain is solved in a rolling manner to obtain the optimal future control sequence, and the first command is issued and executed. See [link to workflow details]. Figure 4 The diagram shown is a schematic of the workflow of a multivariable collaborative real-time optimization controller provided by an embodiment of the present invention.

[0124] Furthermore, in some preferred embodiments of the present invention, the step of using a model predictive control framework to solve a multi-objective optimization problem in each control cycle based on the model-predicted future internal state includes: at the beginning of each control cycle, using the current physical stack state data acquired through a high-speed synchronous data link as the initial state of the high-precision dynamic digital twin model to complete the state synchronization between the virtual model and the physical entity; setting a prediction time domain greater than or equal to the control time domain, and using the high-precision dynamic digital twin model after state synchronization to predict the future state response of the system corresponding to different candidate control sequences applied in the prediction time domain; and solving a multi-objective optimization problem in a finite time domain online to find the control... The optimal control sequence that minimizes the objective function value within the time domain; wherein the objective function includes at least a first weight term for penalizing the deviation between the maximum temperature difference of the fuel cell cross section and the preset target value, a second weight term for penalizing the deviation between the net power of the system and the preset reference value, and a third smoothness weight term for penalizing the change amplitude of the control action; the decision variables of the optimization problem are the set values ​​of hydrogen flow rate, air flow rate, and coolant flow rate at each step in the future control time domain; the solution process of the optimization problem is constrained by preset constraints, which include at least the system dynamic constraints described by the high-precision dynamic digital twin model, the flow rate and speed safety range constraints of each actuator, the rate of change constraints of the control command, and the safety boundary constraints of key state variables.

[0125] Specifically, at the beginning time k of each control cycle, the current physical stack state data collected through the high-speed synchronous data link is used as the initial state x(k) of the high-precision dynamic digital twin model to complete the state synchronization between the virtual model and the physical entity.

[0126] Define a prediction time domain that is greater than or equal to the control time domain. Using the high-precision dynamic digital twin model after state synchronization is completed, predict the future state response x(k+i|k), i=1,...,P of the system corresponding to different candidate control sequences u(k+i|k), i=0,...,M-1 applied in the prediction time domain.

[0127] Solve an online multi-objective optimization problem in a finite-time domain to find the optimal control sequence U(k) = [u(k), u×(k+1), ..., u×(k+M-1)] that minimizes the objective function value in the control time domain. T ;

[0128] The objective function is:

[0129] ;

[0130] in, This represents the maximum temperature difference across the fuel cell stack cross-section. This is a net power reference value; This represents the actual net power. The input vector is the control vector; (k+i) represents the i-th sampling time in the future. Objective function weights: temperature uniformity weight α ranges from 0.5 to 10, and is 5.0 in this embodiment; net power tracking weight β ranges from 0.1 to 5, and is 0.5 in this embodiment; control smoothness weight γ ranges from 0.01 to 2, and is 0.1 in this embodiment.

[0131] The decision variables of the optimization problem are the set values ​​of hydrogen flow rate, air flow rate, and coolant flow rate at each step in the future control time domain; the solution process of the optimization problem is subject to preset constraints, which include at least the following:

[0132] System dynamic constraints: It is described by the high-precision dynamic digital twin model;

[0133] Control constraints: This refers to the safe range of flow rate and speed. For example, the air flow rate (equivalent to the air compressor speed) can be set to 15,000~22,000 rpm, the hydrogen flow rate can be set to 120~190 slpm, and the cooling pump opening can be set to 45%~60%.

[0134] Control Incremental Constraints: To ensure smooth execution;

[0135] State constraints include: Temperature safety upper limit; To prevent single battery undervoltage; Lower limit of stoichiometry; etc.

[0136] Due to model f mpc Typically nonlinear, it is solved in real time using the Sequential Quadratic Programming (SQP) numerical optimization algorithm.

[0137] In this embodiment, the sampling control period Ts is set to 1.0 second, the prediction time domain P is set to 10 seconds, and the control time domain M is set to 5 seconds. By simultaneously incorporating temperature uniformity, optimal efficiency, and control stability into the optimization objectives, and considering the coordination of multiple actuators and multiple constraints, globally optimal scheduling of hydrogen, air, and coolant flow rates is achieved. This solves the drawback of traditional sub-loop control, which suffers from neglecting certain aspects. Under dynamic operating conditions, it can significantly improve the stack temperature uniformity, dynamic response speed, and system net efficiency.

[0138] Furthermore, in some preferred embodiments of the present invention, the maximum temperature difference of the fuel cell stack cross section is determined by post-processing the three-dimensional temperature field data output from the high-precision dynamic digital twin model and calculating the difference between the maximum and minimum temperatures on multiple key cross sections inside the fuel cell stack; the net power of the system is determined by the difference between the output power of the fuel cell stack and the power consumed by all auxiliary systems.

[0139] Specifically, by post-processing the three-dimensional temperature field output by the model, the maximum temperature difference of the stack cross section can be directly obtained, making the internal temperature distribution uniformity index, which was originally difficult to measure online, a control target that can be directly optimized. By accurately calculating the net power of the system, it is ensured that the optimization process is always guided by improving the actual output efficiency, avoiding one-sided optimization that only pursues the power of the stack itself while ignoring parasitic power consumption.

[0140] Furthermore, in some preferred embodiments of the present invention, a multi-objective optimization problem in a finite time domain is solved online using a real-time iterative sequential quadratic programming algorithm. A preset number of iterations are performed in each control cycle to ensure that the solution is completed and the optimal control sequence is output within the preset control cycle time.

[0141] Specifically, a real-time iterative sequential quadratic programming algorithm is adopted and a fixed number of iterations are performed to ensure that the solution of the nonlinear optimization problem can be stably completed and the command output can be completed within a 1-second control cycle. This solves the engineering problem of uncontrollable computation time of high-precision nonlinear models in real-time control and meets the stringent requirements of embedded controllers for computational determinism.

[0142] Step S108: The first instruction in the cooperative control instruction sequence obtained from the current control cycle is sent to the actuator via a high-speed synchronous data link to adjust the reactant flow rate of the physical fuel cell stack in real time.

[0143] Specifically, the first control command u*(k) obtained from the optimization solution is encoded into a command frame that conforms to the actuator communication protocol. It is then sent to actuators such as the air compressor, hydrogen supply proportional valve, and cooling pump through the same deterministic network, and the actual feedback value from the actuators is received to form a closed-loop verification.

[0144] Furthermore, in some preferred embodiments of the present invention, the method further includes: after executing the optimal control command at the current moment, at the beginning of the next control cycle, instead of using the predicted trajectory of the previous cycle, the physical stack state data at the current moment obtained through the high-speed synchronous data link is directly used as the initial state for a new round of rolling optimization.

[0145] Specifically, by directly using the measured state to refresh the prediction starting point in each cycle, a closed-loop feedback correction mechanism is constructed, which can effectively suppress the impact of small model mismatches and external disturbances on control accuracy, enhance the robustness of the control system, and ensure that even if there are prediction deviations, the actual control effect can still maintain stable convergence.

[0146] Furthermore, in some preferred embodiments of the present invention, the method further includes: continuously monitoring the prediction accuracy index of the high-precision dynamic digital twin model during recent operation; when the average value of the sliding window of the prediction accuracy index exceeds a preset warning threshold, automatically triggering a background offline learning process; in the background offline learning process, automatically extracting a representative segment of operating data from the historical database, and using intelligent optimization algorithms to re-optimize the hyperparameters of the online parameter dynamic correction module and some fixed parameters in the kernel of the multiphysics coupling mechanism with the goal of minimizing the comprehensive prediction error of the model on the representative operating data; and safely injecting and replacing the original parameters in the high-precision dynamic digital twin model with a hot update method, without interrupting the normal operation of the system, using the verified re-optimized parameter set.

[0147] Specifically, the prediction accuracy indicators of the high-precision dynamic digital twin model during recent operation are continuously monitored, such as the mean absolute percentage error of temperature prediction; when the average value of the prediction accuracy indicator exceeds the preset warning threshold, the background offline learning process is automatically triggered.

[0148] In the aforementioned offline learning process, a representative segment of running data from the historical database is automatically extracted. With the goal of minimizing the model's overall prediction error on this representative running data, an intelligent optimization algorithm is employed to optimize the hyperparameters (such as the forgetting factor λ and the initial covariance matrix P(0)) of the online parameter dynamic correction module and some fixed parameters (such as the cathode exchange current density i) in the multiphysics coupling mechanism kernel. 0cathode Further optimization is performed using the scaling factor.

[0149] The validated and optimized parameter set is then safely injected into and replaces the original parameters in the high-precision dynamic digital twin model via hot update, without interrupting the normal operation of the system.

[0150] In this embodiment, the prediction accuracy index is the average absolute percentage error of temperature prediction, and the warning threshold is set to 8%. Learning is triggered when the error exceeds 10% for 5 consecutive periods. The background optimization uses an intelligent optimization algorithm (such as particle swarm optimization algorithm) to further optimize the relevant parameters. After optimization, the online model is updated to restore the prediction accuracy to the target level.

[0151] By constructing a system-level intelligent maintenance closed loop covering "precision monitoring - automatic learning - hot update", the control system has the long-term adaptive capability to cope with fuel cell stack performance degradation and environmental changes. This solves the pain point of traditional control systems' performance degradation over time, and maintains high-performance control throughout the entire life cycle of the fuel cell stack, significantly improving the system's reliability and maintainability.

[0152] For example, consider the parameters of the following physical fuel cell system:

[0153] The fuel cell stack has a rated power of 80kW and consists of 350 single cells connected in series. It adopts a parallel flow channel bipolar plate design.

[0154] Air supply subsystem: includes air compressor (with intercooler), hydrogen circulation pump, and front-end humidifier.

[0155] Thermal management subsystem: includes a frequency-controlled main cooling pump, a three-way proportional control valve, a low-temperature radiator, and a PTC heater.

[0156] Sensor networks include:

[0157] Temperature sensors: 24 PT100 platinum resistance thermometers. Four sensors are placed at the coolant inlet, outlet, and cathode flow channel outlets at three cross-sections along the stack length of 25%, 50%, and 75%, for a total of 12 sensors; two sensors are placed at the anode inlet and return manifolds; and four sensors are placed at key cooling pipeline nodes.

[0158] Pressure and flow sensors: air intake manifold (mass flow meter, absolute pressure sensor), hydrogen intake manifold (mass flow meter, absolute pressure sensor), coolant main circuit (electromagnetic flow meter, pressure sensor).

[0159] Voltage monitoring module: 1 set, capable of simultaneously collecting the voltage of all 350 individual batteries, with a sampling frequency ≥10Hz.

[0160] Actuators: air compressor (driven by brushless DC motor), hydrogen supply proportional valve, cooling pump (controlled by frequency converter), three-way valve (driven by servo motor).

[0161] The control system hardware platform includes:

[0162] Main controller: Employs a high-performance industrial real-time controller (such as NI cRIO-9039), runs a real-time operating system, and is responsible for the core computing and system coordination of the MPC algorithm and digital twin model.

[0163] Data acquisition and synchronization unit: Employs EtherCAT-based distributed I / O modules (such as the NI 9144 chassis with analog and digital input / output modules). The main controller acts as the EtherCAT master, and each I / O module and driver acts as a slave.

[0164] Clock synchronization: The system is equipped with a high-precision IEEE 1588 PTP master clock card, which is distributed through the EtherCAT network to achieve sub-microsecond clock synchronization of all I / O nodes and intelligent sensor transmitters.

[0165] Construction and initialization of a high-precision dynamic digital twin model:

[0166] Geometric modeling and mesh generation:

[0167] Based on the CAD drawings of the fuel cell stack, a simplified 3D model containing a representative single cell (including bipolar plates, membrane electrodes, and flow channels) is built in ANSYS Fluent or COMSOL Multiphysics. To balance computational speed and accuracy, an equivalent method of "representative unit + series thermal resistance network" is adopted for the entire stack.

[0168] Unstructured meshes were created for the fluid domain (flow channel) and solid domain (bipolar plate, MEA), with meshes refined near the wall and in the reaction layer region. The total number of meshes was kept below 500,000 to ensure that the model could complete a single forward simulation within seconds.

[0169] Multiphysics model parameter settings:

[0170] Electrochemical model: A quasi-two-dimensional model was used. Membrane electrode parameters (catalyst loading, membrane thickness, etc.) were provided by the supplier. Exchange current density i θ Kinetic parameters such as double-layer capacitance were initially calibrated using electrochemical impedance spectroscopy (EIS).

[0171] Mass transfer and fluid model: The porosity and tortuosity factor of the gas diffusion layer were obtained from the material data sheet. The flow channel friction coefficient f was obtained through computational fluid dynamics (CFD) pre-simulation and fitted as a function of the Reynolds number Re.

[0172] Heat transfer model: The density, specific heat capacity, and thermal conductivity of each material layer (graphite plate, stainless steel, membrane electrode assembly) are based on standard values ​​from the material library. The contact thermal resistance between the cooling channel and the bipolar plate is considered as a parameter to be identified.

[0173] Online parameter dynamic correction module initialization:

[0174] Correction parameter set θ: Initially set as follows:

[0175] .

[0176] Algorithm initialization: Recursive least squares with forgetting factor (FFRLS) is used. The forgetting factor λ is set to 0.995, and the initial parameter covariance matrix P(0) is set to a large diagonal matrix (e.g., 106*I) to indicate large initial uncertainty.

[0177] Observations: Select those that are sensitive to and easy to measure the above parameters.

[0178] ;

[0179] Multivariable Collaborative Real-Time Optimization Controller (MPC) Configuration:

[0180] The sampling and control period Ts is set to 1.0 second. The prediction time domain P = 10 (i.e., predicting the next 10 seconds). The control time domain M = 5 (i.e., optimizing the control action within the next 5 seconds, keeping the control quantity unchanged in the last 5 steps).

[0181] Specific optimization settings:

[0182] Control variables:

[0183] ;

[0184] Objective function weights (after preliminary adjustments): Temperature uniformity weight = 5.0, Net power tracking weight = 0.5 (reference net power Pnet_ref is given by the vehicle controller based on the accelerator pedal request), Control smoothness weight = 0.1.

[0185] Constraints:

[0186] Control constraints:

[0187] , , ;

[0188] Control Incremental Constraints:

[0189] , , ;

[0190] State constraints:

[0191] , , , ;

[0192] Solver configuration:

[0193] C code for this nonlinear optimization problem is generated using real-time optimization software (such as ACADO Toolkit or CasADi).

[0194] The solver uses a real-time iterative sequential quadratic programming (SQP) algorithm. Within each control cycle, the solver performs a fixed number of iterations (e.g., 3 times) to ensure that the calculation is completed and the result is output within a 1-second cycle.

[0195] System integration and operation process:

[0196] After the system is powered on, the high-speed data link starts up, completing the PTP clock synchronization across the entire network.

[0197] The digital twin model loads initial parameters and begins operation after receiving sensor data. The FFRLS correction module then begins working, rapidly correcting the model parameters θ to match the current physical fuel cell stack during the initial few minutes of system warm-up.

[0198] The MPC controller is automatically activated once the digital twin model output is stable and the accuracy assessment meets the standard (MAPET < 5%).

[0199] Steady-state and dynamic coordinated control operation:

[0200] Steady-state cruise (power 40kW):

[0201] The MPC controller, based on the target net power, collaboratively optimizes... =15000rpm, =120slpm =45% of the instructions. The model predicted a maximum temperature difference of 3.5°C for the fuel cell stack, while the actual measured temperature difference was 3.8°C, with MAPET remaining at around 2%.

[0202] Dynamic acceleration process (40kW→70kW, hill climbing requirements):

[0203] At time k: The vehicle controller issues a power step request. The MPC controller immediately uses the current state as a starting point and calls the digital twin model to predict the next 10 seconds. The optimization solution reveals that, in order to quickly increase power while suppressing temperature rise, it is necessary to significantly increase airflow and moderately improve cooling in advance.

[0204] Output instruction: u*(k)=[ =22000rpm (surge) =190slpm (increased synchronously). =60% (pre-cooling enhancement)].

[0205] At times k+1, k+2, ..., the controller executes the commands in a rolling fashion and continuously corrects the predictions using actual temperature and voltage feedback. Throughout the entire 5-second ramp-up process, the maximum actual temperature difference of the fuel cell stack was controlled within 8°C, and no local overheating occurred; the voltage rose steadily without any sudden drop caused by starving.

[0206] Dynamic accuracy guarantee mechanism operation:

[0207] Assume that after the system has been running for several months, the membrane electrode will experience slight aging, and the catalyst activity will decrease.

[0208] Accuracy degradation: On a certain day, under typical cruise conditions, the average value of MAPET's sliding window slowly rose to 9% (exceeding the 8% warning line) and was recorded by the module.

[0209] Learning Triggered: The following day, under similar operating conditions, MAPET exceeded 10% for five consecutive cycles. The dynamic accuracy verification and self-learning module was triggered.

[0210] Self-learning process: The module automatically extracts all relevant data (approximately 200,000 data points) from the past 24 hours. In the background, the PSO optimization algorithm optimizes the initial covariance matrix P(0) of the FFRLS module and the i0cathode scaling factor in the mechanistic model with the goal of minimizing the combined voltage and temperature prediction error on this dataset. After approximately 10 minutes of offline computation, a new parameter set θnew is obtained.

[0211] Hot update and recovery: After being validated with independent historical data, θnew was safely updated to the online model. After the update, MAPET quickly dropped back below 4% in subsequent runs. The model completed an adaptive adjustment to performance degradation, regaining the accuracy requirement of "deviation ≤10%", and the entire process required no manual intervention and did not affect the normal operation of the vehicles.

[0212] On the same 80kW fuel cell system, the performance of the traditional PID split-loop control strategy and the collaborative real-time optimization control strategy provided in this embodiment were compared under dynamic operating conditions. The test cycle included frequent acceleration / deceleration and constant power sections, as shown in Table 1.

[0213] Table 1

[0214]

[0215] Experimental results show that the proposed scheme significantly outperforms traditional PID control in several key performance indicators: during dynamic processes, the maximum temperature difference of the fuel cell stack is reduced from 12-18℃ under traditional PID control to 5-9℃, greatly improving temperature uniformity; the standard deviation of the system net efficiency fluctuation decreases from ±2.1% to ±0.8%, significantly improving operational stability; the load step response time (reaching 90% target power) is shortened from 3.5 seconds to 2.1 seconds, resulting in a faster dynamic response; the minimum single-cell voltage during dynamic processes is increased from 0.58V to 0.65V, effectively avoiding the risk of reactant starvation; and the average absolute percentage error of the temperature prediction of the digital twin model is kept stable between 3% and 7% in the long term, even when the traditional simplified model is not applied or exceeds 15%, fully verifying the high fidelity and long-term stability of the model.

[0216] The technical solution provided by this invention constructs a dynamic digital twin through "multi-physics coupling mechanism kernel + online parameter dynamic correction". While ensuring the accuracy of CFD-electrochemistry-heat transfer coupling, it uses model simplification and recursive least squares (RLS) to achieve second-level online updates, realizing the unification of high-precision model and real-time control. This fundamentally solves the contradiction between high-fidelity simulation and real-time control, and keeps the model prediction error stable within 5% for a long time.

[0217] The technical solution provided by this invention is based on a high-reliability twin model. The MPC controller simultaneously optimizes the flow rates of hydrogen, air, and coolant as multiple variables, and performs rolling solutions with the goal of minimizing temperature difference and maximizing efficiency. This achieves true multi-variable dynamic collaborative optimization, breaking through the limitations of traditional split-loop PID or single-target MPC. Under dynamic operating conditions, it reduces the maximum temperature difference by more than 40% and significantly improves response speed and net efficiency.

[0218] The technical solution provided by this invention forms a real-time adaptive intelligent control system of "perception-modeling-decision-execution," from hardware synchronous data acquisition and deterministic network transmission to online model calibration and controller rolling optimization. The system possesses dynamic precision monitoring and self-learning capabilities, automatically adapting to fuel cell stack performance degradation and achieving continuous optimal control, significantly improving the system's reliability and long-term stability.

[0219] The technical solution provided in this invention constructs a high-fidelity, online-evolvable dynamic digital twin model as the control core. It innovatively integrates a multi-physics coupling mechanism model of computational fluid dynamics, electrochemistry, and heat transfer with an online parameter dynamic correction module based on recursive least squares. This model, through reasonable model dimensionality reduction (such as equivalent flow resistance networks and quasi-two-dimensional electrochemical discretization), meets real-time requirements while ensuring the accuracy of key physical processes. Furthermore, it can dynamically track and correct model parameters (such as exchange current density and membrane conductivity) through sensor data, ensuring that the virtual model always maintains a high-precision mapping (deviation ≤10%) with the physical stack state, providing a reliable predictive basis for real-time optimization.

[0220] The technical solution provided by this invention establishes a multivariate collaborative model predictive control architecture with a high-fidelity twin model as the prediction engine. Unlike traditional MPC which uses a highly simplified linear model, this invention directly integrates and calls the aforementioned high-precision, nonlinear digital twin model as the prediction model for MPC. Under this architecture, the controller can perform multivariate synchronous rolling optimization of hydrogen flow, air flow, and coolant flow based on real physical coupling relationships, directly embedding "minimizing the maximum temperature difference of the fuel cell stack cross-section" into the optimization objective. This achieves deep collaboration from the model level to the control level, fundamentally solving the multivariate mismatch problem under dynamic operating conditions.

[0221] The technical solution provided in this invention presents a system-level intelligent maintenance mechanism with a closed-loop adaptive "control-precision" mechanism. This invention goes beyond general control loops, constructing a system-level protection closed loop covering model precision monitoring, early warning, self-learning, and hot updates. By calculating model prediction errors (such as temperature MAPE) in real time, a background learning process is automatically triggered when precision deteriorates. This process uses intelligent optimization algorithms to recalibrate key parameters and safely and seamlessly update the online model with the new parameters. This mechanism enables the system to have long-term adaptive capabilities to cope with performance degradation and environmental changes, ensuring the durability and robustness of control performance.

[0222] The purpose of this invention is to provide a real-time optimization control method for temperature and flow rate coordination in fuel cell stacks. By constructing a high-speed synchronous data link, microsecond-level synchronous acquisition and low-latency transmission of physical stack operating status data are achieved, providing a precise spatiotemporal data foundation for subsequent modeling and control. By pre-setting a high-precision dynamic digital twin model and dynamically correcting key parameters online, the model can track the state evolution and characteristic drift of the physical stack in real time, always maintaining high-precision synchronization with the physical entity, thereby providing reliable predictive capabilities for control decisions. Based on this high-reliability model, a model predictive control framework is used to solve multi-objective optimization problems in a rolling manner, directly generating a coordinated control command sequence for hydrogen, air, and coolant flow rates. This achieves end-to-end closed-loop optimization from data acquisition to control decision-making, fundamentally solving the problem of the disconnect and difficulty in coordination between temperature and flow control in traditional methods. Under dynamic operating conditions, it can effectively improve the stack temperature uniformity and system efficiency.

[0223] Example 2

[0224] Based on the above embodiments, this invention provides a real-time optimization control system for fuel cell stack temperature-flow rate coordination, see [link to relevant documentation]. Figure 5 The diagram shown is a schematic representation of a real-time optimization control system for temperature-flow coordination in a fuel cell stack, provided by an embodiment of the present invention. The system includes:

[0225] The data acquisition module 310 is used to synchronously acquire real-time operating status data of the physical fuel cell stack via a high-speed synchronous data link.

[0226] The model prediction module 320 is used to input real-time operating status data into a preset high-precision dynamic digital twin model and run the model to predict the future internal state of the physical fuel cell stack in real time. During the operation, based on the deviation between the real-time operating status data and the model prediction value, the key parameters of the model are dynamically corrected online to ensure that the model always maintains high-precision synchronization with the physical fuel cell stack.

[0227] The optimization control module 330 is used to predict the future internal state based on the model. It adopts a model predictive control framework to solve a multi-objective optimization problem in each control cycle and generate a coordinated control command sequence for hydrogen flow, air flow, and coolant flow. The multi-objective optimization problem has at least the optimization objectives of minimizing the internal temperature non-uniformity of the fuel cell stack and optimizing the system efficiency.

[0228] The instruction execution module 340 is used to send the first instruction in the cooperative control instruction sequence obtained by solving the current control cycle to the actuator through a high-speed synchronous data link so as to adjust the reactant flow rate of the physical fuel cell stack in real time.

[0229] Furthermore, in some preferred embodiments of the present invention, the data acquisition module 310 is used to deploy distributed synchronous acquisition nodes in multiple key physical areas of the physical fuel cell stack; wherein, the key physical areas include at least one of the following: cathode flow channel outlet, coolant inlet and outlet, air intake manifold, and multiple representative cross sections along the length of the stack; a global synchronization pulse signal is generated by a master clock source, and all acquisition nodes are triggered to start sampling at the same hardware clock edge through the global synchronization pulse signal, so as to synchronously acquire temperature, pressure, flow rate and voltage data of the key physical areas with microsecond-level precision; each acquisition node preprocesses the acquired raw signal and adds a precise timestamp to obtain a time-stamped data packet; the time-stamped data packet is transmitted to the data service layer through a network transport layer based on deterministic real-time Ethernet.

[0230] Furthermore, in some preferred embodiments of the present invention, the high-precision dynamic digital twin model includes: a multiphysics coupling mechanism kernel; the multiphysics coupling mechanism kernel includes at least an electrochemical sub-model, a computational fluid dynamics sub-model, and an unsteady heat transfer sub-model, and the electrochemical sub-model, the computational fluid dynamics sub-model, and the unsteady heat transfer sub-model are co-calculated using a bidirectional strong coupling method; a model prediction module 320, used to calculate the gas velocity, gas pressure, and gas component concentration distribution in the flow channel based on the current state by the computational fluid dynamics sub-model; and to convert the gas velocity, gas pressure, and gas component concentration distribution into a single value. Gas pressure and gas component concentration distribution are input into the electrochemical sub-model, which calculates the spatially distributed current density and local reaction heat generation rate. The local reaction heat generation rate is then injected as an internal heat source into the unsteady-state heat transfer sub-model, driving it to calculate the three-dimensional transient temperature field inside the stack. The calculated three-dimensional transient temperature field is fed back to the electrochemical sub-model and the computational fluid dynamics sub-model to update the temperature-dependent electrochemical dynamics parameters in the electrochemical sub-model and the temperature-dependent fluid property parameters in the computational fluid dynamics sub-model. The calculation is iterative until convergence.

[0231] Furthermore, in some preferred embodiments of the present invention, the model prediction module 320 is used to execute a recursive closed-loop process of prediction-comparison-update in each control cycle, including: driving the multiphysics coupling mechanism kernel to perform a forward simulation based on the updated key parameter estimates from the previous time step and the input at the current time step, calculating the observed prediction values ​​corresponding to the current time step; comparing the observed prediction values ​​with the real-time sensor observation values ​​obtained through a high-speed synchronous data link at the current time step, and calculating the prediction error vector between the observed prediction values ​​and the real-time sensor observation values; using a recursive least squares algorithm with a forgetting factor, calculating the gain matrix in real time based on the prediction error vector and the pre-calculated sensitivity matrix, and updating the estimates of key parameters based on the gain matrix; wherein, the key parameters include at least: cathode exchange current density and membrane conductivity parameters; and injecting the updated key parameter estimates into the multiphysics coupling mechanism kernel in real time to replace the old values, so that the model can run in the corrected state in the next control cycle.

[0232] Furthermore, in some preferred embodiments of the present invention, the optimization control module 330 is used to, at the beginning of each control cycle, use the current physical stack state data acquired through a high-speed synchronous data link as the initial state of the high-precision dynamic digital twin model to complete the state synchronization between the virtual model and the physical entity; set a prediction time domain greater than or equal to the control time domain, and use the high-precision dynamic digital twin model after state synchronization to predict the future state response of the system corresponding to different candidate control sequences applied in the prediction time domain; and solve a multi-objective optimization problem in a finite time domain online to find the optimal control sequence that minimizes the objective function value in the control time domain; wherein The objective function includes at least a first weight term for penalizing the deviation between the maximum temperature difference of the fuel cell cross section and the preset target value, a second weight term for penalizing the deviation between the net power of the system and the preset reference value, and a third smoothness weight term for penalizing the change amplitude of the control action. The decision variables of the optimization problem are the set values ​​of hydrogen flow rate, air flow rate, and coolant flow rate at each step in the future control time domain. The solution process of the optimization problem is subject to preset constraints, which include at least the system dynamic constraints described by the high-precision dynamic digital twin model, the flow rate and speed safety range constraints of each actuator, the rate of change of control commands, and the safety boundary constraints of key state variables.

[0233] Furthermore, in some preferred embodiments of the present invention, the maximum temperature difference of the fuel cell stack cross section is determined by post-processing the three-dimensional temperature field data output from the high-precision dynamic digital twin model and calculating the difference between the maximum and minimum temperatures on multiple key cross sections inside the fuel cell stack; the net power of the system is determined by the difference between the output power of the fuel cell stack and the power consumed by all auxiliary systems.

[0234] Furthermore, in some preferred embodiments of the present invention, a multi-objective optimization problem in a finite time domain is solved online using a real-time iterative sequential quadratic programming algorithm. A preset number of iterations are performed in each control cycle to ensure that the solution is completed and the optimal control sequence is output within the preset control cycle time.

[0235] Furthermore, in some preferred embodiments of the present invention, the device further includes: a feedback correction module, which, after executing the optimal control command at the current moment, at the beginning of the next control cycle, no longer uses the predicted trajectory of the previous cycle, but directly uses the physical stack state data at the current moment obtained through the high-speed synchronous data link as the initial state for a new round of rolling optimization.

[0236] Furthermore, in some preferred embodiments of the present invention, the device further includes: a self-learning module, used to continuously monitor the prediction accuracy index of the high-precision dynamic digital twin model during recent operation; when the average value of the sliding window of the prediction accuracy index exceeds a preset warning threshold, the background offline learning process is automatically triggered; in the background offline learning process, a representative segment of running data from the historical database is automatically extracted, and with the goal of minimizing the comprehensive prediction error of the model on the representative running data, an intelligent optimization algorithm is used to re-optimize the hyperparameters of the online parameter dynamic correction module and some fixed parameters in the kernel of the multiphysics coupling mechanism; the re-optimized parameter set that has passed verification is safely injected into and replaces the original parameters in the high-precision dynamic digital twin model in a hot update manner without interrupting the normal operation of the system.

[0237] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the fuel cell stack temperature-flow coordinated real-time optimization control system described above can be referred to the corresponding process in the embodiments of the aforementioned fuel cell stack temperature-flow coordinated real-time optimization control method, and will not be repeated here.

[0238] Example 3

[0239] This invention also provides an electronic device for running a real-time optimization control method for temperature-flow coordination in a fuel cell stack; see [link to related documentation]. Figure 6 The schematic diagram of an electronic device provided by the embodiment of the present invention shown includes a memory 400 and a processor 401. The memory 400 is used to store one or more computer instructions, which are executed by the processor 401 to realize the above-mentioned real-time optimization control method for temperature-flow coordination of fuel cell stack.

[0240] Furthermore, Figure 6The electronic device shown also includes a bus 402 and a communication interface 403. The processor 401, the communication interface 403 and the memory 400 are connected via the bus 402.

[0241] The memory 400 may include high-speed random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 403 (which can be wired or wireless), such as the Internet, wide area network, local area network, metropolitan area network, etc. The bus 402 can be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 6 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0242] Processor 401 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of processor 401 or by instructions in software form. Processor 401 can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor. The software module can reside in a readily available storage medium in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory 400, and processor 401 reads information from memory 400 and, in conjunction with its hardware, completes the steps of the method described in the foregoing embodiments.

[0243] This invention also provides a computer-readable storage medium storing computer-executable instructions. When these computer-executable instructions are called and executed by a processor, they cause the processor to implement the aforementioned real-time optimization control method for temperature-flow coordination in a fuel cell stack. For specific implementation details, please refer to the method embodiments, which will not be repeated here.

[0244] The computer program product of the fuel cell stack temperature-flow coordinated real-time optimization control method, system and electronic device provided in the embodiments of the present invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods in the preceding method embodiments. For specific implementation, please refer to the method embodiments, which will not be repeated here.

[0245] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system and / or device described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0246] Furthermore, in the description of the embodiments of the present invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in the present invention based on the specific circumstances.

[0247] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0248] Finally, 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 foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for real-time optimization control of temperature-flow coordinated operation in a fuel cell stack, characterized in that, include: Real-time operating status data of the physical fuel cell stack is collected synchronously via a high-speed synchronous data link. The real-time operating status data is input into a preset high-precision dynamic digital twin model, and the model is run to predict the future internal state of the physical fuel cell stack in real time. During the operation, based on the deviation between the real-time operating status data and the model prediction value, the key parameters of the model are dynamically corrected online to ensure that the model always maintains high-precision synchronization with the physical fuel cell stack. Based on the future internal state predicted by the model, a model predictive control framework is adopted to solve a multi-objective optimization problem in each control cycle, generating a coordinated control command sequence for hydrogen flow rate, air flow rate, and coolant flow rate; wherein, the multi-objective optimization problem has at least the optimization objectives of minimizing the internal temperature non-uniformity of the fuel cell stack and optimizing system efficiency; The first instruction in the cooperative control instruction sequence obtained from the current control cycle is sent to the actuator through the high-speed synchronous data link to adjust the reactant flow rate of the physical fuel cell stack in real time. Based on the model's prediction of the future internal state, a model predictive control framework is employed, and the steps for solving a multi-objective optimization problem on a rolling basis within each control cycle include: At the beginning of each control cycle, the current physical stack state data acquired through the high-speed synchronous data link is used as the initial state of the high-precision dynamic digital twin model to complete the state synchronization between the virtual model and the physical entity. A prediction time domain greater than or equal to the control time domain is defined, and the high-precision dynamic digital twin model after state synchronization is used to predict the future state response of the system corresponding to different candidate control sequences applied in the prediction time domain. An online solution is used to solve a multi-objective optimization problem within a finite time domain to find the optimal control sequence that minimizes the objective function value within the control time domain. The objective function includes at least a first weight term penalizing the deviation between the maximum temperature difference of the fuel cell stack cross-section and a preset target value, a second weight term penalizing the deviation between the net system power and a preset reference value, and a third smoothness weight term penalizing the amplitude of control action changes. The decision variables of the optimization problem are the setpoints for hydrogen flow rate, air flow rate, and coolant flow rate at each step within the future control time domain. The solution process is constrained by preset constraints, which include at least the system dynamics constraints described by the high-precision dynamic digital twin model, the flow rate and rotational speed safety range constraints of each actuator, the rate of change constraints of control commands, and the safety boundary constraints of key state variables. The maximum temperature difference of the fuel cell stack cross-section is determined by post-processing the three-dimensional temperature field data output by the high-precision dynamic digital twin model, calculating the difference between the maximum and minimum temperatures at multiple key cross-sections within the fuel cell stack. The net system power is determined based on the difference between the fuel cell stack output power and the power consumed by all auxiliary systems.

2. The fuel cell stack temperature-flow coordinated real-time optimization control method according to claim 1, characterized in that, The steps for synchronously acquiring real-time operating status data of a physical fuel cell stack via a high-speed synchronous data link include: Distributed synchronous acquisition nodes are deployed in multiple key physical regions of the physical fuel cell stack; wherein, the key physical regions include at least one of the following: cathode flow channel outlet, coolant inlet and outlet, air intake manifold, and multiple representative cross sections along the length of the stack; A global synchronization pulse signal is generated by the master clock source. The global synchronization pulse signal triggers all acquisition nodes to start sampling at the same hardware clock edge, so as to synchronously acquire the temperature, pressure, flow and voltage data of the key physical area with microsecond-level precision. Each acquisition node preprocesses the acquired raw signal and adds a precise timestamp to obtain a time-stamped data packet; The time-stamped data packets are transmitted to the data service layer via a network transport layer based on deterministic real-time Ethernet.

3. The fuel cell stack temperature-flow coordinated real-time optimization control method according to claim 1, characterized in that, The high-precision dynamic digital twin model includes: a multiphysics coupling mechanism kernel; the multiphysics coupling mechanism kernel includes at least an electrochemical sub-model, a computational fluid dynamics sub-model, and an unsteady heat transfer sub-model, and the electrochemical sub-model, computational fluid dynamics sub-model, and unsteady heat transfer sub-model are co-calculated using a bidirectional strong coupling method; after the step of inputting the real-time operating status data into the preset high-precision dynamic digital twin model, the method further includes: The computational fluid dynamics sub-model calculates the gas velocity, gas pressure, and gas component concentration distribution within the flow channel based on the current state; The gas velocity, gas pressure, and gas component concentration distribution are input into the electrochemical sub-model, and the electrochemical sub-model calculates the spatially distributed current density and local reaction heat generation rate. The local reaction heat generation rate is injected into the unsteady heat transfer sub-model as an internal heat source, driving the unsteady heat transfer sub-model to calculate the three-dimensional transient temperature field inside the stack. The calculated three-dimensional transient temperature field is fed back to the electrochemical sub-model and the computational fluid dynamics sub-model to update the temperature-dependent electrochemical dynamic parameters in the electrochemical sub-model and the temperature-dependent fluid property parameters in the computational fluid dynamics sub-model. The calculation is iterated until convergence.

4. The fuel cell stack temperature-flow coordinated real-time optimization control method according to claim 3, characterized in that, The steps of dynamically correcting the key parameters of the model online based on the deviation between the real-time running status data and the model's predicted values ​​during operation include: Within each control cycle, a recursive closed-loop process of prediction-comparison-update is executed, including: Based on the key parameter estimates updated at the previous time step and the input at the current time step, the multiphysics coupling mechanism kernel is driven to perform a forward simulation to calculate the observation prediction value corresponding to the current time step. The predicted observation value is compared with the real-time sensor observation value obtained through the high-speed synchronous data link at the current moment, and the prediction error vector between the predicted observation value and the real-time sensor observation value is calculated. A recursive least squares algorithm with a forgetting factor is used to calculate the gain matrix in real time based on the prediction error vector and the pre-calculated sensitivity matrix, and to update the estimated values ​​of the key parameters based on the gain matrix; wherein the key parameters include at least: cathode exchange current density and membrane conductivity parameters. The updated estimates of the key parameters are injected into the multiphysics coupling mechanism kernel in real time to replace the old values, enabling the model to run in the corrected state in the next control cycle.

5. The fuel cell stack temperature-flow coordinated real-time optimization control method according to claim 1, characterized in that, The online solution to a multi-objective optimization problem in a finite time domain employs a real-time iterative sequential quadratic programming algorithm. A preset number of iterations are performed within each control cycle to ensure that the solution is completed and the optimal control sequence is output within the preset control cycle time.

6. The fuel cell stack temperature-flow coordinated real-time optimization control method according to claim 1, characterized in that, The method further includes: After executing the optimal control command for the current moment, at the start of the next control cycle, the predicted trajectory of the previous cycle is no longer used. Instead, the physical stack state data of the current moment, obtained through the high-speed synchronous data link, is used directly as the initial state for a new round of rolling optimization.

7. The fuel cell stack temperature-flow coordinated real-time optimization control method according to claim 1, characterized in that, The method further includes: The prediction accuracy index of the high-precision dynamic digital twin model is continuously monitored during recent operation. When the average value of the prediction accuracy index during the sliding window exceeds the preset warning threshold, the background offline learning process is automatically triggered. In the background offline learning process, a representative segment of running data is automatically extracted from the historical database. With the goal of minimizing the comprehensive prediction error of the model on the representative running data, an intelligent optimization algorithm is used to further optimize the hyperparameters of the online parameter dynamic correction module and some fixed parameters in the kernel of the multiphysics coupling mechanism. The validated and optimized parameter set is then safely injected into and replaces the original parameters in the high-precision dynamic digital twin model via hot update, without interrupting the normal operation of the system.

8. A real-time optimization control system for temperature-flow coordination in a fuel cell stack, characterized in that, include: The data acquisition module is used to synchronously acquire real-time operating status data of the physical fuel cell stack via a high-speed synchronous data link; The model prediction module is used to input the real-time operating status data into a preset high-precision dynamic digital twin model, and run the model to predict the future internal state of the physical fuel cell stack in real time. During the operation, based on the deviation between the real-time operating status data and the model prediction value, the key parameters of the model are dynamically corrected online to ensure that the model always maintains high-precision synchronization with the physical fuel cell stack. An optimization control module is used to solve a multi-objective optimization problem in each control cycle based on the future internal state predicted by the model, using a model predictive control framework, and generating a coordinated control command sequence for hydrogen flow rate, air flow rate, and coolant flow rate; wherein, the multi-objective optimization problem has at least the optimization objectives of minimizing the internal temperature non-uniformity of the fuel cell stack and optimizing system efficiency; The instruction execution module is used to send the first instruction in the cooperative control instruction sequence obtained by solving the current control cycle to the actuator through the high-speed synchronous data link, so as to adjust the reactant flow rate of the physical fuel cell stack in real time. The optimization control module is used at the beginning of each control cycle to use the current physical stack state data acquired through the high-speed synchronous data link as the initial state of the high-precision dynamic digital twin model, thereby synchronizing the virtual model with the physical entity's state. It sets a prediction time domain greater than or equal to the control time domain and uses the synchronized high-precision dynamic digital twin model to predict the future system state response corresponding to different candidate control sequences applied within the prediction time domain. It then solves an online multi-objective optimization problem within a finite time domain to find the optimal control sequence that minimizes the objective function value within the control time domain. The objective function includes at least a first weight term to penalize the deviation between the maximum temperature difference of the stack cross-section and a preset target value, and a weight term to penalize the deviation between the system net power and a preset reference value. The second weight term and the third smoothness weight term used to penalize changes in the control action amplitude; the decision variables of the optimization problem are the set values ​​of hydrogen flow rate, air flow rate, and coolant flow rate at each step in the future control time domain; the solution process of the optimization problem is constrained by preset constraints, which include at least the system dynamics constraints described by the high-precision dynamic digital twin model, the flow rate and speed safety range constraints of each actuator, the rate of change constraints of control commands, and the safety boundary constraints of key state variables; the maximum temperature difference of the fuel cell stack cross section is determined by post-processing the three-dimensional temperature field data output by the high-precision dynamic digital twin model and calculating the difference between the maximum and minimum temperatures on multiple key cross sections inside the fuel cell stack; the net power of the system is determined by the difference between the output power of the fuel cell stack and the power consumed by all auxiliary systems.

Images