Low-temperature start-up hydrothermal co-control method for automotive fuel cells

Abstract

This invention relates to the field of fuel cell thermal management technology, specifically a method for coordinated water and heat control during low-temperature startup of automotive fuel cells. The method includes: collecting multi-source operating data on stack temperature distribution, membrane electrode humidity, and reactant gas pressure during the initial stage of low-temperature startup; employing a fuzzy neural network algorithm optimized based on the phase change kinetics of water in the low-temperature membrane electrode and the change law of proton conductivity to calculate initial heating and humidification strategies; and generating coordinated control commands for heater power, gas humidification, and back pressure valve opening based on real-time assessment results of the risk of water freezing inside the stack. The commands are executed while monitoring the stack voltage uniformity and the rate of voltage drop of individual cells, dynamically optimizing the control commands until the stack temperature and internal humidity reach preset operating standards. This method enables precise coordinated regulation of the water and heat state during low-temperature startup, suppressing water freezing inside the stack, stabilizing the stack voltage operating state, and smoothly completing the low-temperature startup process of the fuel cell.

Description

Technical Field

[0001] This invention relates to the field of fuel cell thermal management technology, and in particular to a method for low-temperature start-up water-thermal coordinated control of automotive fuel cells. Background Technology

[0002] Under the low-temperature start-up conditions of automotive fuel cells, existing conventional control methods only collect a single type of operating parameter and use a general fuzzy neural network algorithm to generate heating and humidification control schemes. The algorithm model has not been adapted and optimized by combining the phase change dynamics of membrane electrode water and the change law of proton conductivity under low-temperature conditions. The output of the control strategy relies on fixed algorithm logic and cannot match the actual internal operating condition evolution characteristics of the fuel cell stack during the low-temperature start-up stage.

[0003] In existing technologies, heating regulation, reactant gas humidification, and back pressure valve opening adjustment all employ independent control modes, failing to achieve coordinated linkage across multiple control dimensions. Furthermore, the lack of a real-time assessment mechanism for the risk of water freezing inside the fuel cell stack prevents the matching of control parameters based on the internal freezing risk status. Throughout the low-temperature startup process, there is no continuous monitoring of the stack voltage uniformity or the rate of voltage drop in individual cells. Once control commands are set, they cannot be dynamically adjusted according to the real-time operating status of the stack, easily leading to an imbalance in internal temperature and humidity, resulting in abnormal phase change and freezing of moisture, and disordered fluctuations in the stack's operating voltage. Existing control modes are insufficient for precise regulation of the hydrothermal state throughout the low-temperature startup process and cannot adapt to the complex operating conditions of the fuel cell stack. A novel hydrothermal coordinated control method is needed to meet the operational requirements of low-temperature startup for automotive fuel cells. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the existing technology and propose a water-thermal coordinated control method for low-temperature start-up of automotive fuel cells.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a low-temperature start-up water-thermal coordinated control method for vehicle fuel cells, comprising:

[0006] During the initial low-temperature startup of the fuel cell, multi-source operating condition data including the internal temperature distribution of the stack, the humidity status of the membrane electrode, and the pressure of the reactant gas were collected.

[0007] Based on the multi-source operating data, a preliminary heating and humidification strategy is calculated using an improved fuzzy neural network control algorithm. The improved fuzzy neural network control algorithm is optimized based on the phase transition dynamics of water in the membrane electrode and the change law of proton conductivity at low temperatures.

[0008] Based on the preliminary heating and humidification strategies, and combined with the real-time assessment results of the risk of water freezing inside the fuel cell stack, a coordinated control command is generated, which includes heater power allocation, reactant gas humidification control, and back pressure valve opening adjustment.

[0009] The coordinated control command is executed, and the stack voltage uniformity and single-cell voltage drop rate are monitored simultaneously. During the startup process, the coordinated control command is dynamically adjusted until the fuel cell stack temperature reaches the preset stable operating threshold and the internal humidity reaches the predetermined range.

[0010] The low-temperature startup refers to the startup process where the initial temperature of the fuel cell stack is between -20°C and 0°C.

[0011] As a further aspect of the present invention, the improved fuzzy neural network control algorithm is optimized based on the dynamic characteristics of water phase transition in membrane electrodes at low temperatures and the variation law of proton conductivity, including:

[0012] A multi-layer feedforward fuzzy neural network is constructed, with its input layer nodes corresponding to key physical quantities in the multi-source operating condition data, and its output layer nodes corresponding to the quantification parameters of the preliminary heating and humidification strategies.

[0013] In the process of generating the membership function of the multilayer feedforward fuzzy neural network, the critical condition of water phase transition inside the membrane electrode at low temperature is introduced as a constraint. The critical condition of water phase transition includes the relationship between supercooling and ice crystal nucleation rate.

[0014] In the network weight adjustment rules of the multilayer feedforward fuzzy neural network, the functional relationship between proton conductivity and membrane water content and temperature is incorporated, so that the network output tends to maintain proton conductivity above the minimum start-up requirement.

[0015] The multilayer feedforward fuzzy neural network is trained using historical low-temperature startup data, and its connection weights and membership function parameters are optimized. This enables the multilayer feedforward fuzzy neural network to map heating and humidification strategies that can effectively suppress intramembrane icing and promote proton conduction based on real-time multi-source operating condition data.

[0016] As a further aspect of the present invention, the step of training the multilayer feedforward fuzzy neural network using historical low-temperature startup data to optimize its connection weights and membership function parameters includes the following steps:

[0017] Collect multi-source operating condition data sequences, executed heating and humidification strategy sequences, and corresponding startup result evaluation data recorded during historical low-temperature startup processes;

[0018] From the historical low-temperature startup data, data samples that successfully started and had optimal water management were extracted to form a positive sample set, while data samples that failed to start due to membrane icing or flooding were extracted to form a negative sample set.

[0019] The multi-source operating condition data in the positive and negative sample sets are used as the input features of the multilayer feedforward fuzzy neural network, and the quantization parameters of the corresponding heating and humidification strategies are used as the expected output.

[0020] Using the backpropagation algorithm, the mean square error between the network output and the expected output is used as the loss function to iteratively update the connection weights and membership function parameters of the multilayer feedforward fuzzy neural network.

[0021] In each iteration, the gradient of the loss function with respect to the network connection weights and membership function parameters is calculated, and the parameter values ​​are adjusted along the gradient descent direction according to the preset learning rate until the loss function converges to the preset threshold or reaches the maximum number of iterations.

[0022] The trained multilayer feedforward fuzzy neural network is deployed in the controller to map heating and humidification strategies that can effectively suppress intramembrane icing and promote proton conduction based on real-time input multi-source operating data.

[0023] As a further aspect of the present invention, the acquisition of multi-source operating condition data, including the internal temperature distribution of the fuel cell stack, the humidity status of the membrane electrode, and the pressure of the reactant gas, includes:

[0024] By arranging temperature sensor arrays at the inlet, outlet and middle of the bipolar plate flow channel of the fuel cell stack, real-time temperature measurement values ​​of different regions of the fuel cell stack are obtained, forming the internal temperature distribution of the fuel cell stack.

[0025] Indirect measurement signals of local water content in the membrane electrode are obtained by a humidity sensor embedded in the edge of the membrane electrode or the flow field plate, and converted into the humidity status of the membrane electrode by a calibration model.

[0026] By installing pressure sensors on the anode intake manifold, cathode intake manifold and exhaust manifold, the absolute pressure and differential pressure data of the reaction gas at key nodes are obtained, forming a set of reaction gas pressure data.

[0027] The real-time temperature measurement value, the indirect measurement signal, and the absolute pressure and differential pressure data are collected synchronously, and aligned and packaged according to a unified timestamp to form the multi-source operating condition data.

[0028] As a further aspect of the present invention, the generation of real-time assessment results of the risk of water freezing inside the fuel cell stack includes:

[0029] Based on the internal temperature distribution of the fuel cell stack, the region with the lowest temperature and the temperature value of the region with the lowest temperature are identified.

[0030] Based on the humidity status of the membrane electrode and the saturated water vapor partial pressure data at the current temperature, calculate the conditions for the existence of liquid water inside the membrane electrode and estimate the amount of water.

[0031] Based on the temperature value of the lowest temperature region, the existence conditions of the liquid water and the estimated water volume, combined with the freezing point and supercooled water thermodynamic data, the real-time risk probability of water undergoing phase change and freezing in the lowest temperature region is calculated.

[0032] The real-time risk probability is compared with a preset risk threshold, and the real-time assessment result of the risk of water freezing inside the fuel cell stack is output. The assessment result includes the risk level and the risk location identifier.

[0033] As a further aspect of the present invention, the generation of the cooperative control command includes:

[0034] The preliminary heating strategy was analyzed to obtain the recommended heating power distribution for different regions of the fuel cell stack.

[0035] Analyze the preliminary humidification strategy to obtain the recommended humidification amount and humidity setpoint for the reactant gas;

[0036] Based on the risk level and risk location marker in the real-time assessment results of the risk of water freezing inside the fuel cell stack, the heating power distribution is corrected, and compensating heating power is added to the area corresponding to the risk location marker.

[0037] Meanwhile, the recommended humidification amount and humidity setting value of the reaction gas are adjusted according to the risk level. When the risk level is high, the humidification amount setting value is reduced to reduce water injection, and when the risk level is low, the humidification amount setting value is maintained or appropriately increased to ensure humidification.

[0038] Based on the back pressure requirements of the fuel cell stack operation and the real-time assessment results of the water freezing risk, the opening adjustment of the back pressure valve is calculated to control the water vapor partial pressure inside the fuel cell stack.

[0039] The corrected heating power distribution, the adjusted setpoint for the humidification of the reaction gas, and the adjustment amount of the back pressure valve opening are encapsulated into the coordinated control command.

[0040] As a further aspect of the present invention, the heating power distribution is modified by adding compensatory heating power to the area corresponding to the risk location marker, including:

[0041] Based on the internal temperature distribution of the fuel cell stack, calculate the temperature difference between the area corresponding to the risk location marker and the average temperature of the fuel cell stack;

[0042] Based on the temperature difference and the real-time risk probability, the magnitude of the compensation heating power is determined through a preset compensation power lookup table. The larger the temperature difference and the higher the risk probability, the greater the compensation heating power.

[0043] The compensated heating power is superimposed on the heating power suggested for the area corresponding to the risk location in the preliminary heating strategy to form the total heating power for the area corresponding to the risk location.

[0044] Check whether the total heating power exceeds the maximum allowable power of the heater in the area corresponding to the risk location marker. If it does, limit the power and redistribute the excess power demand to the adjacent area heater according to a preset rule.

[0045] As a further aspect of the present invention, the synchronous monitoring of stack voltage uniformity and single-cell voltage drop rate, and the dynamic adjustment of the cooperative control commands during startup, include:

[0046] During startup, the real-time voltage of each cell in the fuel cell stack is continuously collected;

[0047] Calculate the standard deviation or the difference between the maximum and minimum values ​​of the real-time voltage of all individual cells as an indicator of stack voltage uniformity.

[0048] Identify the single cell with the lowest voltage and calculate its voltage change rate over time as the single cell voltage drop rate.

[0049] When the voltage uniformity index of the fuel cell stack exceeds the first set threshold, or the voltage drop rate of a single cell exceeds the second set threshold, it is determined that the hydrothermal management is unbalanced or there is a local abnormality.

[0050] Based on the determination result of the imbalance or local anomaly in the water and heat management, a dynamic adjustment process for the collaborative control command is triggered. The dynamic adjustment process includes re-collecting the multi-source operating data and generating and executing the updated collaborative control command based on the latest multi-source operating data and the real-time assessment result of the water freezing risk.

[0051] As a further aspect of the present invention, the dynamic adjustment process includes:

[0052] Suspend the execution of the current collaborative control command;

[0053] Immediately re-collect the multi-source operating condition data, and re-perform the real-time assessment of the risk of water freezing inside the fuel cell stack based on the latest multi-source operating condition data to generate the latest real-time assessment result of water freezing risk.

[0054] The latest multi-source operating condition data and the latest real-time assessment results of water freezing risk are input again into the improved fuzzy neural network control algorithm to calculate the updated preliminary heating and humidification strategies.

[0055] Based on the updated preliminary heating and humidification strategies, and combined with the latest real-time assessment results of water freezing risk, updated collaborative control instructions are generated.

[0056] Execute the updated cooperative control instructions and continue the steps of synchronously monitoring the stack voltage uniformity and the single-cell voltage drop rate.

[0057] As a further aspect of the present invention, the method further includes an exit determination step for the low-temperature start-up process:

[0058] Continuously monitor the temperature and membrane electrode humidity of the fuel cell stack;

[0059] When the lowest temperature of the fuel cell stack reaches the preset stable operating threshold, and the time during which the humidity state of the membrane electrode remains within the predetermined range exceeds the preset stable duration, the low-temperature start-up is determined to be successfully completed.

[0060] Generate a start-up completion signal and switch to the water and thermal management strategy under normal fuel cell operation.

[0061] Record key data sequences throughout the entire process from startup to startup completion. These key data sequences include multi-source operating condition data sequences, cooperative control command sequences, and stack voltage uniformity index sequences, which are used to update the training dataset of the improved fuzzy neural network control algorithm.

[0062] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0063] The dynamic characteristics of the water phase transition in the membrane electrode assembly (MEA) at low temperatures and the variation law of proton conductivity are integrated into the optimization process of the fuzzy neural network control algorithm. Based on the optimized algorithm, the heating and humidification strategies for the initial stage of low-temperature startup are calculated and solved. The algorithm architecture closely matches the inherent physical changes of the MEA under low-temperature conditions in fuel cells, mitigating the adaptation bias of general control algorithms in low-temperature scenarios. The control parameters based on multi-source operating condition data output can match the actual operating conditions of temperature, humidity, and gas pressure inside the fuel cell stack, ensuring that the initial control logic aligns with the physical property changes within the fuel cell stack during the low-temperature startup phase.

[0064] By incorporating real-time assessment results of the risk of water freezing inside the fuel cell stack, the power distribution of the heater, the control of the humidification of the reactant gas, and the adjustment of the back pressure valve opening are integrated and coordinated to form an integrated collaborative control command system. Using the uniformity of the stack voltage and the rate of voltage drop of individual cells as monitoring criteria, the collaborative control commands are dynamically corrected throughout the entire process, breaking the independent operation mode of each control unit and achieving synchronous and linked changes in multi-dimensional control parameters. The control process can continuously adapt and adjust according to the real-time operating status of the fuel cell stack. The changes in internal temperature and humidity of the stack maintain a synchronous rhythm, the internal water phase change process tends to be stable, the development trend of abnormal phase change in water under low-temperature conditions is weakened, and all operating parameters of the fuel cell stack can smoothly transition to the preset stable operating range. Attached Figure Description

[0065] Figure 1 The flowchart is a process for the low-temperature start-up water-thermal coordinated control method for vehicle fuel cells according to the present invention;

[0066] Figure 2 A flowchart illustrating the work on the optimization method for the improved fuzzy neural network control algorithm;

[0067] Figure 3 This is a flowchart illustrating the working process of a multi-source operating condition data acquisition method. Detailed Implementation

[0068] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0069] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the 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 the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0070] See Figure 1 This invention provides a method for low-temperature start-up water-thermal coordinated control of automotive fuel cells, the overall implementation of which is as follows:

[0071] During the initial low-temperature startup of the fuel cell, the system collects multi-source operating condition data, including the internal temperature distribution of the fuel cell stack, the humidity status of the membrane electrode assembly (MEA), and the pressure of the reactant gas. Based on the collected multi-source operating condition data, an improved fuzzy neural network control algorithm is used to calculate preliminary heating and humidification strategies. This improved fuzzy neural network control algorithm is optimized based on the phase transition kinetics of water in the MEA at low temperatures and the change law of proton conductivity. Subsequently, based on the calculated preliminary heating and humidification strategies and combined with the real-time assessment results of the risk of water freezing inside the fuel cell stack, the system generates coordinated control commands, including heater power allocation, reactant gas humidification control, and back pressure valve opening adjustment. The coordinated control commands are issued and executed. During the startup process of executing the commands, the system simultaneously monitors the voltage uniformity of the fuel cell stack and the rate of voltage drop of individual cells. When the monitored indicators are abnormal, the executed coordinated control commands are dynamically adjusted. This process is repeated until the fuel cell stack temperature reaches the preset stable operating threshold and the internal humidity status reaches the predetermined range.

[0072] In one embodiment of the present invention, see [reference] Figure 2 A multilayer feedforward fuzzy neural network was constructed, with input layer nodes corresponding to key physical quantities in multi-source operating data and output layer nodes corresponding to the quantification parameters of preliminary heating and humidification strategies. During the membership function generation process of this multilayer feedforward fuzzy neural network, a critical condition for water phase transition inside the membrane electrode at low temperatures was introduced as a constraint. This critical condition includes the relationship between supercooling and ice crystal nucleation rate. The network weight adjustment rules of this multilayer feedforward fuzzy neural network incorporate the functional relationship between proton conductivity and membrane water content and temperature, making the network output tend to maintain proton conductivity above the minimum start-up requirement.

[0073] The multilayer feedforward fuzzy neural network (MLN) was trained using historical low-temperature startup data to optimize its connection weights and membership function parameters. The training process included the following steps: collecting multi-source operating condition data sequences, executed heating and humidification strategy sequences, and corresponding startup result evaluation data recorded during historical low-temperature startups. From the historical low-temperature startup data, data samples showing successful startups and optimal water management were extracted to form a positive sample set, while data samples showing startup failures due to membrane freezing or flooding were extracted to form a negative sample set. The multi-source operating condition data from both the positive and negative sample sets were used as input features to the MNN, and the quantized parameters of the corresponding heating and humidification strategies were used as the desired output. Using the backpropagation algorithm, with the mean squared error between the network output and the desired output as the loss function, the connection weights and membership function parameters of the MNN were iteratively updated. In each iteration, the gradient of the loss function relative to the network connection weights and membership function parameters was calculated, and the parameter values ​​were adjusted along the gradient descent direction according to a preset learning rate until the loss function converged to a preset threshold or the maximum number of iterations was reached. The trained multilayer feedforward fuzzy neural network is deployed in the controller to map heating and humidification strategies that can effectively suppress intramembrane icing and promote proton conduction based on real-time multi-source operating data.

[0074] In practical implementation, the construction of a multilayer feedforward fuzzy neural network (MLNN) is a core step. The input layer nodes of the MLNN correspond to key physical quantities in the multi-source operating data, including the minimum and average temperatures in the internal temperature distribution of the fuel cell stack, the average water content in the membrane electrode humidity state, and the cathode inlet pressure and pressure difference in the reaction gas pressure. The output layer nodes of the MLNN correspond to the quantification parameters of the initial heating and humidification strategies, including the power setpoints for each zone heater and the setpoint for the humidifier output humidity. During the generation of the membership function of the MLNN, a critical condition for the water phase transition inside the membrane electrode at low temperatures is introduced as a constraint. This critical condition includes the relationship between supercooling and ice crystal nucleation rate. This makes the generated membership function more sensitive to the shape of the input variable "membrane electrode humidity state" within the range of high supercooling. In the network weight adjustment rules of the multilayer feedforward fuzzy neural network, the functional relationship between proton conductivity and membrane water content and temperature is incorporated. The functional relationship is embedded in the calculation process of network output error in analytical form, so that when the network adjusts the weights, its gradient descent direction is always optimized in the direction of maintaining proton conductivity above the minimum start-up requirement.

[0075] In some embodiments, a multilayer feedforward fuzzy neural network is trained using historical low-temperature startup data to optimize its connection weights and membership function parameters. Multi-source operating condition data sequences, executed heating and humidification strategy sequences, and corresponding startup result evaluation data recorded during historical low-temperature startup are collected. From the historical low-temperature startup data, data samples with successful startup and optimal water management are extracted to form a positive sample set, while data samples with startup failures due to membrane icing or flooding are extracted to form a negative sample set. The multi-source operating condition data from the positive and negative sample sets are used as input features to the multilayer feedforward fuzzy neural network, and the quantized parameters of the corresponding heating and humidification strategies are used as the desired output. An error backpropagation algorithm is used, with the mean square error between the network output and the desired output as the loss function, to iteratively update the connection weights and membership function parameters of the multilayer feedforward fuzzy neural network. In each iteration, the gradient of the loss function relative to the network connection weights and membership function parameters is calculated, and the parameter values ​​are adjusted along the gradient descent direction according to a preset learning rate until the loss function converges to a preset threshold or reaches the maximum number of iterations. The trained multilayer feedforward fuzzy neural network is deployed in the controller to map heating and humidification strategies that can effectively suppress intramembrane icing and promote proton conduction based on real-time multi-source operating data.

[0076] Optionally, the critical condition constraint for water phase change introduced during the membership function generation process can be mathematically expressed as an adjustment factor related to supercooling. This adjustment factor acts on the membership function width parameter of the humidity input variable. Above, the expression is:

[0077]

[0078] in: Represents supercooling. It is a monotonically increasing function fitted based on experimental data of ice crystal nucleation rate. This relationship means that when the supercooling of the membrane electrode environment increases, the membership function width decreases, and the network responds more quickly to subtle changes in humidity input, thus triggering heating or humidification reduction strategies earlier in the control logic. It can be understood that this approach directly encodes the physical laws of water phase transition into the structural initialization of the fuzzy neural network, rather than relying entirely on data-driven learning.

[0079] In some embodiments, the proton conductivity variation function incorporated into the network weight adjustment rule directly affects the calculation of gradient components in the error backpropagation algorithm. Specifically, when calculating the network output layer error, not only is the difference between the quantization parameters of the heating and humidification strategies of the network output and the expected output calculated, but a penalty term based on the proton conductivity function is also introduced. This penalty term is associated with the intramembrane proton conductivity value predicted by the current network output strategy. When the predicted proton conductivity value is lower than the minimum startup requirement, the penalty term increases the overall loss function value. This means that the error backpropagation algorithm, when adjusting weights, is simultaneously guided by two objectives: "strategy approximation accuracy" and "maintaining proton conduction capability," resulting in a network with better robustness under low-temperature startup conditions. Optionally, the historical low-temperature startup data used for updating needs to undergo rigorous preprocessing, including data alignment, filtering and denoising, and normalization, to ensure the stability and convergence of the multilayer feedforward fuzzy neural network training.

[0080] In one embodiment of the present invention, see [reference] Figure 3 A temperature sensor array positioned at the inlet, outlet, and middle of the bipolar plate flow channel of the fuel cell stack acquires real-time temperature measurements of different regions of the stack, forming an internal temperature distribution. Humidity sensors embedded at specific locations on the membrane electrode edge or flow field plate acquire indirect measurement signals of local water content at the membrane electrode, which are then converted into membrane electrode humidity status using a calibration model. Pressure sensors installed on the anode inlet manifold, cathode inlet manifold, and exhaust manifold acquire absolute pressure and differential pressure data of the reactant gas at key nodes, forming a reactant gas pressure data set. The real-time temperature measurements, indirect measurement signals, and absolute pressure and differential pressure data are simultaneously acquired, aligned, and packaged according to a unified timestamp to constitute the multi-source operating condition data.

[0081] The process of generating a real-time assessment result of the risk of water freezing inside the fuel cell stack includes: identifying the lowest temperature region and its temperature value based on the internal temperature distribution of the fuel cell stack; calculating the conditions for the presence of liquid water inside the membrane electrode and the estimated water volume based on the humidity status of the membrane electrode and the saturated water vapor partial pressure data at the current temperature; calculating the real-time risk probability of water undergoing phase change freezing in the lowest temperature region based on the temperature value, liquid water presence conditions, and estimated water volume, combined with freezing point and supercooled water thermodynamic data; comparing the real-time risk probability with a preset risk threshold, and outputting a real-time assessment result of the risk of water freezing inside the fuel cell stack, which includes the risk level and risk location identifier.

[0082] In practical implementation, the acquisition of multi-source operating data is accomplished through a sensor network deployed on the fuel cell stack. Temperature sensor arrays, positioned at the inlet, outlet, and middle of the bipolar plate flow channels, acquire real-time temperature measurements of different regions of the stack. The arrangement of the temperature sensor array ensures coverage of the main reaction areas and edge regions of the stack, thus forming an internal temperature distribution that reflects spatial differences. Humidity sensors embedded at specific locations on the membrane electrode edge or flow field plate acquire indirect measurement signals of local water content at the membrane electrode. These indirect measurement signals are typically impedance or resistance signals, which are converted into membrane electrode humidity status using a pre-established calibration model. The calibration model describes the correspondence between impedance signals and intramembrane water content under specific temperature and pressure conditions. Pressure sensors installed on the anode inlet manifold, cathode inlet manifold, and exhaust manifold acquire absolute pressure and differential pressure data of the reactant gases at key nodes, forming a set of reactant gas pressure data including inlet pressure, exhaust pressure, and flow channel pressure drop. The data acquisition unit synchronously acquires real-time temperature measurements from the temperature sensor array, indirect measurement signals from the humidity sensor, and absolute pressure and differential pressure data from the pressure sensor. These data are then aligned and packaged according to a unified timestamp to form a time-synchronized multi-source operating condition data packet.

[0083] The generation of a real-time assessment result of the risk of water freezing inside the fuel cell stack is a computational process. Based on the internal temperature distribution of the stack, the lowest temperature region and its temperature value are identified. This lowest temperature region can be one or more specific stack sections or individual cell locations. Based on the membrane electrode humidity state and combined with saturated water vapor partial pressure data at the current temperature, the conditions for the presence of liquid water inside the membrane electrode and the estimated water volume are calculated. This calculation is based on the gas partial pressure law and the membrane water balance model. Based on the temperature value of the lowest temperature region, the conditions for the presence of liquid water, and the estimated water volume, combined with freezing point and supercooled water thermodynamic data, the real-time risk probability of water undergoing phase change and freezing within the lowest temperature region is calculated. The calculated real-time risk probability is compared with a preset risk threshold, and the real-time assessment result of the risk of water freezing inside the stack is output. The assessment result includes a risk level and a risk location identifier. The risk level can be divided into three levels: "low," "medium," and "high," and the risk location identifier points to the physical coordinates of the lowest temperature region.

[0084] In some embodiments, a comprehensive quantitative model can be used to calculate the real-time risk probability. Optionally, one implementation uses temperature and estimated liquid water volume as key input parameters, the relationship of which can be expressed as:

[0085]

[0086] in: Indicates the real-time risk probability. This indicates the temperature value of the identified lowest temperature region. This represents the estimated amount of liquid water calculated based on the humidity state and gas conditions of the membrane electrode in the lowest temperature region. The function is a mapping function constructed based on thermodynamic phase diagrams and nucleation theory; its characteristic is that the lower the temperature and the higher the temperature, the more... The larger the calculated output value, the better. This formula combines multiple influencing factors into a single probability value, facilitating threshold judgment in subsequent control logic.

[0087] In some embodiments, the calculation of the membrane electrode humidity state involves signal conversion. Indirect measurement signals acquired by humidity sensors embedded at specific locations on the membrane electrode edge or flow field plate need to be converted using a calibration model. Optionally, the calibration model can be a set of two-dimensional lookup tables obtained from calibration experiments at different temperatures and pressures. The real-time impedance signal, along with the simultaneously acquired local temperature and pressure values, are used as inputs, and the corresponding membrane electrode water content value is output through interpolation. It is understood that the accuracy of this method depends on whether the preliminary calibration experiments adequately cover the temperature, pressure, and water content range that may be encountered during the low-temperature start-up of the fuel cell. Calculating the conditions for the presence of liquid water inside the membrane electrode requires comparing the calculated water content value with the saturated water content at the current temperature and pressure. If the former is greater than the latter, liquid water is considered to be present, and the portion exceeding the saturation value can be estimated as the predicted water volume based on the membrane electrode pore volume.

[0088] In one embodiment of the present invention, the process of generating cooperative control commands includes: parsing a preliminary heating strategy calculated by an improved fuzzy neural network control algorithm to obtain the suggested heating power distribution for different regions of the fuel cell stack; parsing a preliminary humidification strategy to obtain the suggested humidification amount and humidity setpoint for the reactant gases; correcting the heating power distribution based on the risk level and risk location marker in the real-time assessment results of the risk of water freezing inside the fuel cell stack, and increasing the compensating heating power for the regions corresponding to the risk location markers; simultaneously adjusting the suggested humidification amount and humidity setpoint for the reactant gases according to the risk level, reducing the humidification setpoint to reduce water injection when the risk level is high, and maintaining or appropriately increasing the humidification setpoint to ensure humidification when the risk level is low; calculating the back pressure valve opening adjustment amount by combining the basic back pressure requirements for fuel cell stack operation and the real-time assessment results of water freezing risk to control the water vapor partial pressure inside the fuel cell stack; and encapsulating the corrected heating power distribution, the adjusted reactant gas humidification setpoint, and the back pressure valve opening adjustment amount into cooperative control commands.

[0089] The specific process of correcting the heating power distribution and increasing compensatory heating power for areas corresponding to risk location markers includes: calculating the temperature difference between the area corresponding to the risk location marker and the average temperature of the fuel cell stack based on the internal temperature distribution. Based on this temperature difference and the real-time risk probability, the magnitude of the compensatory heating power is determined using a preset compensatory power lookup table; the larger the temperature difference and the higher the risk probability, the greater the compensatory heating power. The determined compensatory heating power is then added to the heating power suggested for the area corresponding to the risk location marker in the initial heating strategy to form the total heating power for that area. The total heating power is checked to see if it exceeds the maximum allowable power of the heater in that area. If it does, a limit is applied, and the excess power demand is redistributed to heaters in adjacent areas according to preset rules.

[0090] In practical implementation, the generation of coordinated control commands is an integrated decision-making process. The preliminary heating strategy, calculated by an improved fuzzy neural network control algorithm, is analyzed to obtain suggested heating power distributions for different regions of the fuel cell stack. These distributions are typically represented as vectors, with each element corresponding to a suggested heating power value for a specific zone of the stack. The preliminary humidification strategy is also analyzed to obtain suggested humidification amounts and humidity setpoints for the reactant gases. These setpoints typically include target dew point temperatures or relative humidity commands for the cathode and anode inlet humidifiers. Based on the real-time assessment of water freezing risk within the stack, including risk level and location markers, the heating power distribution in the preliminary heating strategy is revised, increasing compensatory heating power for areas corresponding to risk location markers. Simultaneously, the suggested humidification amounts and humidity setpoints for the reactant gases are adjusted based on the risk level in the real-time water freezing risk assessment. When the risk level is high, the humidification setpoint is reduced to decrease water injection; when the risk level is low, the humidification setpoint is maintained or appropriately increased to ensure humidification. By combining the basic back pressure requirements for fuel cell stack operation with the real-time assessment results of water freezing risk, the opening adjustment of the back pressure valve is calculated to control the water vapor partial pressure inside the fuel cell stack, thereby indirectly affecting the phase change conditions of water. The corrected heating power distribution, the adjusted setpoint for reactant gas humidification, and the opening adjustment of the back pressure valve are encapsulated into a structured, parseable coordinated control command that can be parsed by the underlying actuators.

[0091] The process of correcting the heating power distribution and adding compensatory heating power to the areas corresponding to risk location markers requires specific calculations. Based on the internal temperature distribution of the fuel cell stack, the temperature difference between the area corresponding to the risk location marker and the average temperature of the fuel cell stack is calculated. The formula for calculating the temperature difference is:

[0092]

[0093] in: Indicates temperature difference. This represents the average temperature of the fuel cell stack, calculated based on the internal temperature distribution within the stack. This indicates the latest temperature measurement value for the area corresponding to the risk location marker. It represents the real-time risk probability based on the calculated temperature difference and real-time assessment of water freezing risk. The compensation heating power is determined by querying a pre-set compensation power lookup table. The larger the temperature difference and the higher the real-time risk probability, the greater the determined compensation heating power. This determined compensation heating power is then added to the heating power suggested for the area corresponding to the risk location marker in the initial heating strategy, forming the total heating power for the area corresponding to the risk location marker. Check whether the total heating power exceeds the maximum allowable power of the heater in the area corresponding to the risk location marker. If the limit is exceeded, a limit will be applied, and it will be set to... The overflow power demand will be redistributed to adjacent area heaters according to preset rules.

[0094] In some embodiments, a preset compensation power lookup table is stored in the controller memory in the form of a two-dimensional table. The row index of the table represents a discrete temperature difference range, the column index represents a discrete real-time risk probability level, and the corresponding compensation heating power value is stored in the table cell. See Table 1 for an example fragment of the compensation power lookup table.

[0095] Table 1: Compensated Heating Power Lookup Table

[0096] Temperature difference range (°C) Risk level: Low Risk level: Medium Risk level: High 0<≤2 0W 50W 100W 2<≤5 30W 100W 200W 5<≤10 100W 200W 350W

[0097] It is understandable that the specific values ​​in the compensation power lookup table are obtained based on the thermal model of the fuel cell stack and the calibration of the low-temperature start-up experiment. Its core logic is to apply stronger compensation heating to areas with large temperature differences and high risk of freezing in order to quickly eliminate local low temperature points.

[0098] Optionally, the redistribution of excess power demand follows the principle of heat diffusion. One implementation involves allocating power proportionally to all zone heaters directly adjacent to the area corresponding to the risk location marker. For example, if an area has four adjacent areas, the power increase for each adjacent zone heater is... Before allocation, it is also necessary to check whether the increased power of heaters in adjacent areas exceeds their own maximum allowable power. This limiting and redistribution mechanism ensures that local heating needs are met to the greatest extent possible within hardware safety constraints, while indirectly improving the temperature environment of the target area by heating adjacent areas.

[0099] In some embodiments, the specific strategy for adjusting the humidification setpoint of the reactant gas is also strictly linked to the risk level. When the real-time assessment result of the water freezing risk shows a risk level of "high," the system will multiply the humidity setpoint given by the initial humidification strategy by a decay coefficient less than 1. (For example, 0.6), thereby reducing the humidification setting value. When the risk level is "low", the humidity setting value given by the preliminary humidification strategy is directly adopted, or multiplied by a coefficient slightly greater than 1. (e.g., 1.1) to appropriately increase humidity. Optionally, the calculation of the back pressure valve opening adjustment will refer to an opening mapping function with the risk level and current internal pressure as inputs, and the function will output an opening increment or a target opening value.

[0100] In one embodiment of the present invention, during startup, the real-time voltage of each cell in the fuel cell stack is continuously collected. The standard deviation or the difference between the maximum and minimum values ​​of the real-time voltages of all cells is calculated as a fuel cell stack voltage uniformity index. The cell with the lowest voltage is identified, and its voltage change rate over time is calculated as the cell voltage drop rate. When the fuel cell stack voltage uniformity index exceeds a first preset threshold, or the cell voltage drop rate exceeds a second preset threshold, it is determined that there is an imbalance in hydrothermal management or a local anomaly. Based on the determination of hydrothermal management imbalance or local anomaly, a dynamic adjustment process for the coordinated control command is triggered. This dynamic adjustment process includes re-collecting multi-source operating condition data and generating and executing updated coordinated control commands based on the latest multi-source operating condition data and the real-time assessment results of water freezing risk.

[0101] The specific steps of the dynamic adjustment process include: pausing the execution of the current collaborative control commands; immediately re-acquiring multi-source operating condition data and re-performing the real-time assessment of the risk of water freezing inside the fuel cell stack based on the latest multi-source operating condition data, generating the latest real-time assessment result of water freezing risk; inputting the latest multi-source operating condition data and the latest real-time assessment result of water freezing risk back into the improved fuzzy neural network control algorithm to calculate the updated preliminary heating and humidification strategies; generating updated collaborative control commands based on the updated preliminary heating and humidification strategies and the latest real-time assessment result of water freezing risk; executing the updated collaborative control commands and continuing to synchronously monitor the stack voltage uniformity and the rate of voltage drop of individual cells.

[0102] In practical implementation, synchronous monitoring of stack voltage uniformity and individual cell voltage drop rate is a continuous activity during startup. During startup, the system continuously collects the real-time voltage of each cell in the stack. The voltage acquisition module scans the voltage signals of all cells at a fixed frequency and calculates the standard deviation or the difference between the maximum and minimum values ​​of the real-time voltages of all cells, which serves as an indicator of stack voltage uniformity. The formula for calculating the stack voltage uniformity indicator is as follows:

[0103]

[0104] in: This indicates the standard deviation of the voltage uniformity of the fuel cell stack. Indicates the total number of individual batteries. This represents the real-time voltage of the first single cell. This represents the average real-time voltage of all individual cells. The cell with the lowest voltage is identified, and its voltage change rate over time is calculated as the cell voltage drop rate. The cell voltage drop rate is obtained by dividing the difference between the voltage value of the lowest-voltage cell in the current sampling period and its voltage value in the previous sampling period by the sampling time interval. When the stack voltage uniformity index exceeds the first set threshold... If the voltage drop rate of a single chip exceeds the second preset threshold, the system determines that there is an imbalance in water and heat management or a local anomaly. Based on the determination of water and heat management imbalance or local anomaly, a dynamic adjustment process for the coordinated control command is triggered. The dynamic adjustment process includes re-collecting multi-source operating data and generating and executing updated coordinated control commands based on the latest multi-source operating data and the real-time assessment of water freezing risk.

[0105] The dynamic adjustment process involves a series of sequential operations. Upon initiation, the current collaborative control commands are first paused, and all actuators maintain their current state or enter a safety hold mode. Immediately, multi-source operating condition data is re-acquired, and based on the latest data, a real-time assessment of the risk of water freezing inside the fuel cell stack is re-executed, generating the latest real-time assessment result. The latest multi-source operating condition data and the latest real-time assessment result are then input again into the improved fuzzy neural network control algorithm to calculate updated preliminary heating and humidification strategies. Based on these updated strategies and combined with the latest assessment, updated collaborative control commands are generated. The updated commands are then executed, and the steps of synchronously monitoring stack voltage uniformity and individual cell voltage drop rate continue, forming a closed-loop control.

[0106] In some embodiments, the values ​​of the first and second set thresholds are not fixed. Optionally, the first and second set thresholds can be dynamically adjusted based on the current average voltage or temperature of the fuel cell stack. One implementation involves establishing a threshold lookup table, see Table 2.

[0107] Table 2: Dynamic Threshold Table for Voltage Monitoring

[0108] Average voltage range of fuel cell stack (V) First set threshold (V) Second set threshold (V / s) 0.10 0.05 0.08 0.03 0.05 0.02

[0109] It is understandable that during the initial startup phase when the voltage is low, the allowable voltage unevenness and rate of decline can be slightly larger; as the voltage increases, the threshold tightens to pursue a smoother startup process. In some embodiments, after determining that there is an imbalance in hydrothermal management or a local anomaly, a rapid compensation action can be performed before triggering the full dynamic adjustment process. For example, if the location of the single cell with the lowest voltage is identified to coincide with the risk location marker in the real-time assessment result of water freezing risk, a short-duration maximum power heating pulse of preset duration is immediately sent to the heater corresponding to that location before recalculating the strategy. It is understood that this rapid compensation can buy time for subsequent strategy recalculation and prevent the situation from deteriorating during the calculation. Optionally, "pausing the execution of the current cooperative control command" in the dynamic adjustment process does not mean that all control outputs are reset to zero, but rather that the execution of subsequent steps in the cooperative control command sequence calculated based on the previous cycle is stopped. The actuator will usually remain in the last effective control state until an updated cooperative control command is received.

[0110] In one embodiment of the present invention, the temperature and membrane electrode assembly (MEA) humidity of the fuel cell stack are continuously monitored. When the minimum temperature of the stack reaches a preset stable operating threshold, and the MEA humidity remains within a predetermined range for a period exceeding a preset stable duration, the low-temperature start-up is deemed successfully completed. A start-up completion signal is generated, and the hydrothermal management strategy is switched to the normal operating state of the fuel cell. Key data sequences are recorded throughout the entire process from start-up to start-up completion. These key data sequences include multi-source operating condition data sequences, cooperative control command sequences, and stack voltage uniformity index sequences, which are used to update the training dataset of the improved fuzzy neural network control algorithm.

[0111] In practical implementation, the exit determination step of the low-temperature start-up process is independent of the main control loop operation, continuously monitoring the temperature of the fuel cell stack and the humidity status of the membrane electrode assembly (MEA). The monitoring process is performed at a fixed sampling period. When the minimum temperature of the stack reaches a preset stable operating threshold, and the MEA humidity status remains within a predetermined range for a period exceeding a preset stable duration, the system determines that the low-temperature start-up has been successfully completed. The low-temperature start-up refers to the start-up process when the initial temperature of the stack is between -20℃ and 0℃. This temperature range covers typical start-up conditions for automotive fuel cells in cold environments, including cold starts under extremely cold conditions and warm starts under near-normal temperatures. Within this temperature range, the phase change kinetics of water inside the MEA are characterized by easy freezing of liquid water and a significant decrease in proton conductivity, requiring a specialized hydrothermal synergistic control strategy to ensure a high start-up success rate. The minimum stack temperature is extracted from real-time collected internal temperature distribution data of the stack, and the MEA humidity status is obtained from real-time collected and converted MEA humidity status data. The preset stable operating threshold is a temperature value, for example... The predetermined range is a range related to the humidity of the membrane electrode, for example, the water content of the membrane electrode meets the following conditions. The preset stable duration is a time value, such as seconds. The decision condition can be described by a logical expression:

[0112]

[0113] in: That is the lowest temperature of the fuel cell stack. and These are the minimum and maximum values ​​of the membrane electrode humidity state, respectively. It refers to the duration during which both temperature and humidity conditions simultaneously meet the requirements. It is the preset stable duration. It is the preset stable operating threshold temperature.

[0114] Upon successful cold start, the system generates a start-up completion signal. This signal triggers the control strategy switching logic, switching the hydrothermal management control from the cold start mode to the hydrothermal management strategy under normal fuel cell operation. The hydrothermal management strategy under normal operation employs closed-loop control based on load demand, which differs from the coordinated heating and humidification control used in the cold start phase.

[0115] In some embodiments, a key data sequence is recorded throughout the entire process from startup to startup completion. This key data sequence is stored in non-volatile memory as a time-series array. The key data sequence includes a multi-source operating condition data sequence, a cooperative control command sequence, and a stack voltage uniformity index sequence. The multi-source operating condition data sequence contains data packets of the stack's internal temperature distribution, membrane electrode humidity status, and reactant gas pressure at each sampling time. The cooperative control command sequence contains heater power allocation commands, reactant gas humidification control commands, and back pressure valve opening adjustment commands issued in each control cycle. The stack voltage uniformity index sequence contains stack voltage uniformity index values ​​calculated for each sampling cycle, such as voltage standard deviation. It is understandable that the data sequences recorded here constitute a complete archive of the cryogenic startup process.

[0116] Optionally, when recording key data sequences, the system generates a separate data file for each cryogenic startup process, named with a timestamp and startup initial conditions (such as initial temperature). These data files are used as the training dataset for subsequent updates to the improved fuzzy neural network control algorithm. When updating the training dataset, the multi-source operating condition data and corresponding cooperative control commands from the data sequence of a new successful startup are added as new training sample pairs to the historical cryogenic startup dataset. In some embodiments, after determining a successful startup, the system also performs a brief integrity self-check to check whether the recorded key data sequences are complete, whether there are any abnormal breakpoints, and generates a data recording quality identifier. This ensures the quality of the data used for algorithm updates and avoids incomplete or abnormal data contaminating the training set. Optionally, the preset stabilization duration can be fine-tuned based on the ambient temperature or the initial startup temperature; for example, at lower initial temperatures, it can be appropriately extended to ensure sufficient stabilization of the stack's internal state.

[0117] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for low-temperature start-up water-thermal coordinated control of automotive fuel cells, characterized in that, include: During the initial low-temperature startup of the fuel cell, multi-source operating condition data including the internal temperature distribution of the stack, the humidity status of the membrane electrode, and the pressure of the reactant gas were collected. Based on the multi-source operating data, a preliminary heating and humidification strategy is calculated using an improved fuzzy neural network control algorithm. The improved fuzzy neural network control algorithm is optimized based on the phase transition dynamics of water in the membrane electrode and the change law of proton conductivity at low temperatures. Based on the preliminary heating and humidification strategies, and combined with the real-time assessment results of the risk of water freezing inside the fuel cell stack, a coordinated control command is generated, which includes heater power allocation, reactant gas humidification control, and back pressure valve opening adjustment. The coordinated control command is executed, and the stack voltage uniformity and single-cell voltage drop rate are monitored simultaneously. During the startup process, the coordinated control command is dynamically adjusted until the fuel cell stack temperature reaches the preset stable operating threshold and the internal humidity reaches the predetermined range. The low-temperature startup refers to the startup process where the initial temperature of the fuel cell stack is between -20°C and 0°C. The improved fuzzy neural network control algorithm is optimized based on the dynamic characteristics of water phase transition in membrane electrodes at low temperatures and the variation law of proton conductivity, including: A multi-layer feedforward fuzzy neural network is constructed, with its input layer nodes corresponding to key physical quantities in the multi-source operating condition data, and its output layer nodes corresponding to the quantification parameters of the preliminary heating and humidification strategies. In the process of generating the membership function of the multilayer feedforward fuzzy neural network, the critical condition of water phase transition inside the membrane electrode at low temperature is introduced as a constraint. The critical condition of water phase transition includes the relationship between supercooling and ice crystal nucleation rate. In the network weight adjustment rules of the multilayer feedforward fuzzy neural network, the functional relationship between proton conductivity and membrane water content and temperature is incorporated, so that the network output tends to maintain proton conductivity above the minimum start-up requirement. The multilayer feedforward fuzzy neural network is trained using historical low-temperature startup data, and its connection weights and membership function parameters are optimized. This enables the multilayer feedforward fuzzy neural network to map heating and humidification strategies that can effectively suppress intramembrane icing and promote proton conduction based on real-time multi-source operating condition data.

2. The method for low-temperature start-up water-thermal coordinated control of vehicle fuel cells according to claim 1, characterized in that, The step of training the multilayer feedforward fuzzy neural network using historical low-temperature startup data to optimize its connection weights and membership function parameters includes the following steps: Collect multi-source operating condition data sequences, executed heating and humidification strategy sequences, and corresponding startup result evaluation data recorded during historical low-temperature startup processes; From the historical low-temperature startup data, data samples that successfully started and had optimal water management were extracted to form a positive sample set, while data samples that failed to start due to membrane icing or flooding were extracted to form a negative sample set. The multi-source operating condition data in the positive and negative sample sets are used as the input features of the multilayer feedforward fuzzy neural network, and the quantization parameters of the corresponding heating and humidification strategies are used as the expected output. Using the backpropagation algorithm, the mean square error between the network output and the expected output is used as the loss function to iteratively update the connection weights and membership function parameters of the multilayer feedforward fuzzy neural network. In each iteration, the gradient of the loss function with respect to the network connection weights and membership function parameters is calculated, and the parameter values ​​are adjusted along the gradient descent direction according to the preset learning rate until the loss function converges to the preset threshold or reaches the maximum number of iterations. The trained multilayer feedforward fuzzy neural network is deployed in the controller to map heating and humidification strategies that can effectively suppress intramembrane icing and promote proton conduction based on real-time input multi-source operating data.

3. The method for low-temperature start-up water-thermal coordinated control of vehicle fuel cells according to claim 1, characterized in that, The data collected includes multi-source operating condition data such as the internal temperature distribution of the fuel cell stack, the humidity status of the membrane electrode, and the pressure of the reactant gas, including: By arranging temperature sensor arrays at the inlet, outlet and middle of the bipolar plate flow channel of the fuel cell stack, real-time temperature measurement values ​​of different regions of the fuel cell stack are obtained, forming the internal temperature distribution of the fuel cell stack. Indirect measurement signals of local water content in the membrane electrode are obtained by a humidity sensor embedded in the edge of the membrane electrode or the flow field plate, and converted into the humidity status of the membrane electrode by a calibration model. By installing pressure sensors on the anode intake manifold, cathode intake manifold and exhaust manifold, the absolute pressure and differential pressure data of the reaction gas at key nodes are obtained, forming a set of reaction gas pressure data. The real-time temperature measurement value, the indirect measurement signal, and the absolute pressure and differential pressure data are collected synchronously, and aligned and packaged according to a unified timestamp to form the multi-source operating condition data.

4. The method for low-temperature start-up water-thermal coordinated control of vehicle fuel cells according to claim 1, characterized in that, The generation of real-time assessment results of the risk of water freezing inside the fuel cell stack includes: Based on the internal temperature distribution of the fuel cell stack, the region with the lowest temperature and the temperature value of the region with the lowest temperature are identified. Based on the humidity status of the membrane electrode and the saturated water vapor partial pressure data at the current temperature, calculate the conditions for the existence of liquid water inside the membrane electrode and estimate the amount of water. Based on the temperature value of the lowest temperature region, the existence conditions of the liquid water and the estimated water volume, combined with the freezing point and supercooled water thermodynamic data, the real-time risk probability of water undergoing phase change and freezing in the lowest temperature region is calculated. The real-time risk probability is compared with a preset risk threshold, and the real-time assessment result of the risk of water freezing inside the fuel cell stack is output. The assessment result includes the risk level and the risk location identifier.

5. The method for low-temperature start-up water-thermal coordinated control of vehicle fuel cells according to claim 4, characterized in that, The generation of the coordinated control commands includes: The preliminary heating strategy was analyzed to obtain the recommended heating power distribution for different regions of the fuel cell stack. Analyze the preliminary humidification strategy to obtain the recommended humidification amount and humidity setpoint for the reactant gas; Based on the risk level and risk location marker in the real-time assessment results of the risk of water freezing inside the fuel cell stack, the heating power distribution is corrected, and compensating heating power is added to the area corresponding to the risk location marker. Meanwhile, the recommended humidification amount and humidity setting value of the reaction gas are adjusted according to the risk level. When the risk level is high, the humidification amount setting value is reduced to reduce water injection, and when the risk level is low, the humidification amount setting value is maintained or appropriately increased to ensure humidification. Based on the back pressure requirements of the fuel cell stack operation and the real-time assessment results of the water freezing risk, the opening adjustment of the back pressure valve is calculated to control the water vapor partial pressure inside the fuel cell stack. The corrected heating power distribution, the adjusted setpoint for the humidification of the reaction gas, and the adjustment amount of the back pressure valve opening are encapsulated into the coordinated control command.

6. The method for low-temperature start-up water-thermal coordinated control of vehicle fuel cells according to claim 5, characterized in that, The heating power distribution is corrected by adding compensatory heating power to the areas corresponding to the risk location markers, including: Based on the internal temperature distribution of the fuel cell stack, calculate the temperature difference between the area corresponding to the risk location marker and the average temperature of the fuel cell stack; Based on the temperature difference and the real-time risk probability, the magnitude of the compensation heating power is determined through a preset compensation power lookup table. The larger the temperature difference and the higher the risk probability, the greater the compensation heating power. The compensated heating power is superimposed on the heating power suggested for the area corresponding to the risk location in the preliminary heating strategy to form the total heating power for the area corresponding to the risk location. Check whether the total heating power exceeds the maximum allowable power of the heater in the area corresponding to the risk location marker. If it does, limit the power and redistribute the excess power demand to the adjacent area heater according to a preset rule.

7. The method for low-temperature start-up water-thermal coordinated control of vehicle fuel cells according to claim 1, characterized in that, The synchronous monitoring of stack voltage uniformity and individual cell voltage drop rate, and the dynamic adjustment of the coordinated control commands during startup, include: During startup, the real-time voltage of each cell in the fuel cell stack is continuously collected; Calculate the standard deviation or the difference between the maximum and minimum values ​​of the real-time voltage of all individual cells as an indicator of stack voltage uniformity. Identify the single cell with the lowest voltage and calculate its voltage change rate over time as the single cell voltage drop rate. When the voltage uniformity index of the fuel cell stack exceeds the first set threshold, or the voltage drop rate of a single cell exceeds the second set threshold, it is determined that the hydrothermal management is unbalanced or there is a local abnormality. Based on the determination result of the imbalance or local anomaly in the water and heat management, a dynamic adjustment process for the collaborative control command is triggered. The dynamic adjustment process includes re-collecting the multi-source operating data and generating and executing the updated collaborative control command based on the latest multi-source operating data and the real-time assessment result of the water freezing risk.

8. The method for low-temperature start-up water-thermal coordinated control of a vehicle fuel cell according to claim 7, characterized in that, The dynamic adjustment process includes: Suspend the execution of the current collaborative control command; Immediately re-collect the multi-source operating condition data, and re-perform the real-time assessment of the risk of water freezing inside the fuel cell stack based on the latest multi-source operating condition data to generate the latest real-time assessment result of water freezing risk. The latest multi-source operating condition data and the latest real-time assessment results of water freezing risk are input again into the improved fuzzy neural network control algorithm to calculate the updated preliminary heating and humidification strategies. Based on the updated preliminary heating and humidification strategies, and combined with the latest real-time assessment results of water freezing risk, updated collaborative control instructions are generated. Execute the updated cooperative control instructions and continue the steps of synchronously monitoring the stack voltage uniformity and the single-cell voltage drop rate.

9. The method for low-temperature start-up water-thermal coordinated control of vehicle fuel cells according to claim 1, characterized in that, The method also includes an exit determination step for the low-temperature startup process: Continuously monitor the temperature and membrane electrode humidity of the fuel cell stack; When the lowest temperature of the fuel cell stack reaches the preset stable operating threshold, and the time during which the humidity state of the membrane electrode remains within the predetermined range exceeds the preset stable duration, the low-temperature start-up is determined to be successfully completed. Generate a start-up completion signal and switch to the water and thermal management strategy under normal fuel cell operation. Record key data sequences throughout the entire process from startup to startup completion. These key data sequences include multi-source operating condition data sequences, cooperative control command sequences, and stack voltage uniformity index sequences, which are used to update the training dataset of the improved fuzzy neural network control algorithm.

Images