Water-cooled proton exchange membrane fuel cell thermal management control system

Abstract

The application relates to the technical field of new energy battery thermal management, in particular to a water-cooled proton exchange membrane fuel cell thermal management control system, which comprises a data acquisition module, a model calculation module, a risk prediction module and a collaborative control module. The data acquisition module collects operating condition data such as stack output current and voltage, cooling liquid temperature and flow, and environmental temperature and humidity; the model calculation module constructs an improved electrochemical-thermal coupling model based on an extended equivalent circuit model, calculates the internal heat generation power distribution and temperature field of the stack; the risk prediction module predicts the cooling liquid phase change risk area and evaluates the local hot spot generation probability; and the collaborative control module adopts an improved fuzzy neural network algorithm optimized by an online self-adaptive mechanism, and generates a cooling liquid circulating pump and stack fan collaborative control instruction in combination with multiple parameters. The scheme can finely analyze the internal thermal state of the stack, accurately predict thermal hazards, and realize dynamic collaborative regulation of the heat dissipation components.

Description

Technical Field

[0001] This invention relates to the field of thermal management technology for new energy batteries, and in particular to a thermal management and control system for water-cooled proton exchange membrane fuel cells. Background Technology

[0002] Water-cooled proton exchange membrane fuel cells are crucial equipment in the field of new energy power generation. Conventional thermal management control systems only collect basic operating parameters of the fuel cell stack and conduct thermal state analysis based on basic equivalent circuit models or simplified linear fitting models. This only provides a rough estimate of the external temperature of the stack and lacks refined analytical methods for understanding the internal heat generation patterns and the global temperature field distribution. Traditional thermal management control modes often rely on fixed thresholds to perform passive control operations, failing to anticipate potential phase change risks in the coolant flow channels or quantify the likelihood of localized hotspots. Furthermore, heat dissipation components often employ individual control modes, resulting in mismatched and unadapted operating parameters among components.

[0003] Conventional electrochemical thermal analysis models have a simple architecture and cannot fully account for the coupling relationship between the electrochemical operating characteristics and thermal conduction of fuel cells, making them unsuitable for the precise calculation of internal thermal parameters of the fuel cell stack under complex operating conditions. Traditional fuzzy neural network algorithms have fixed structures and parameters, lacking dynamic adaptive adjustment capabilities. They cannot combine coolant phase change risks, hotspot generation probabilities, and coolant temperature and flow parameters to generate control commands that adapt to the linkage of multiple components in real time, making them unsuitable for the dynamic adjustment requirements of thermal management under varying operating conditions of fuel cells. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the existing technology and propose a thermal management and control system for a water-cooled proton exchange membrane fuel cell.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a thermal management control system based on a water-cooled proton exchange membrane fuel cell, comprising:

[0006] The data acquisition module acquires the operating condition data of the fuel cell stack, including the stack output current, stack output voltage, coolant inlet temperature, coolant outlet temperature, coolant flow rate, ambient temperature, and ambient humidity.

[0007] The model calculation module calculates the heat generation power distribution and temperature field inside the fuel cell stack based on the operating condition data using an improved electrochemical-thermal coupling model. The improved electrochemical-thermal coupling model is constructed based on an extended equivalent circuit model.

[0008] The risk prediction module predicts the phase change risk area of ​​the coolant inside the water-cooled channel based on the heat generation power distribution and temperature field inside the fuel cell stack, and assesses the probability of local hot spot generation.

[0009] The collaborative control module employs an improved fuzzy neural network algorithm to generate collaborative control commands for the coolant circulation pump and the fuel cell stack fan based on the phase transition risk region, the probability of local hotspot generation, the coolant inlet temperature, the coolant outlet temperature, and the coolant flow rate. The improved fuzzy neural network algorithm is optimized based on an online adaptive mechanism.

[0010] As a further aspect of the present invention, the improved electrochemical-thermal coupling model includes the following steps:

[0011] Using the output current and output voltage of the fuel cell stack, the real-time ohmic internal resistance, activation internal resistance, and concentration internal resistance of the fuel cell stack are calculated, and an extended equivalent circuit model containing the mapping relationship between the three internal resistance parameters and the current density is established.

[0012] Based on the extended equivalent circuit model, the energy loss caused by activation polarization, ohmic polarization and concentration polarization during the operation of the fuel cell stack is calculated, and this energy loss is converted into the real-time heat source intensity of each region of the fuel cell stack.

[0013] A three-dimensional computational mesh is generated inside the fuel cell stack along the gas flow path and the membrane electrode stack direction.

[0014] A three-dimensional heat transfer network model was established, including a solid structure, a membrane electrode, a gas diffusion layer, a bipolar plate, and cooling channels.

[0015] The intensity of the real-time heat source is used as an internal heat source and loaded into the corresponding grid position in the three-dimensional heat transfer network model;

[0016] The coolant inlet temperature, coolant outlet temperature, coolant flow rate, ambient temperature, and ambient humidity are used as boundary conditions for the three-dimensional heat transfer network model.

[0017] Solving the three-dimensional heat transfer network model yields the temperature distribution of each solid component and fluid region inside the fuel cell stack, i.e., the heat generation power distribution and temperature field inside the fuel cell stack.

[0018] As a further aspect of the present invention, the predicted phase change risk region of the coolant inside the water-cooled flow channel includes:

[0019] Extract the temperature distribution of the cooling channel region in the three-dimensional heat transfer network model, and identify the temperature value of the inner wall of the cooling channel and its gradient along the flow direction;

[0020] Obtain the working pressure of the coolant at the inlet of the cooling channel, and based on the coolant inlet temperature, query the saturation temperature of the coolant at the working pressure;

[0021] The temperature value of the inner wall of the cooling channel is compared with the saturation temperature point by point. If the wall temperature is higher than the saturation temperature, the corresponding location is determined to be a potential phase change risk point.

[0022] For all identified potential phase change risk points, calculate their temperature change rate along the flow channel axis. If the temperature change rate exceeds the threshold, the potential phase change risk point is determined to have the risk of local boiling leading to bubble accumulation, and is marked as the phase change risk area.

[0023] As a further aspect of the present invention, the evaluation of the probability of local hotspot generation includes:

[0024] The temperature distribution of the three-dimensional heat transfer network model in the membrane electrode, catalyst layer and proton exchange membrane regions is extracted to obtain the local temperature peaks and their spatial coordinates in the membrane electrode, catalyst layer and proton exchange membrane regions;

[0025] The difference between the local temperature peak and the volume-weighted average temperature of the entire fuel cell stack is calculated to obtain the local superheat.

[0026] The risk level of local temperature runaway is assessed based on the local overheating and the temperature distribution uniformity index of the membrane electrode, catalyst layer, and proton exchange membrane regions.

[0027] Based on the risk level of the local temperature runaway and combined with the fluctuation trend of the current stack output current, the quantitative probability of irreversible high-temperature damage to the membrane electrode in the next control cycle is calculated, that is, the probability of local hot spot generation.

[0028] As a further aspect of the present invention, the improved fuzzy neural network algorithm includes the following steps:

[0029] The area ratio of the phase transition risk region, the probability of local hotspot generation, the coolant inlet temperature, the coolant outlet temperature, the coolant flow rate, the ambient temperature, and the ambient humidity are used as input variables for the input layer.

[0030] The input variables are preprocessed by fuzzification, and each input variable is transformed into the membership value of multiple fuzzy subsets through multiple fuzzy membership functions;

[0031] The membership value is input into a multilayer feedforward neural network, which contains at least one hidden layer, and the network weights are set according to expert experience in the initial stage.

[0032] The output of the multilayer feedforward neural network is the desired speed of the coolant circulation pump and the desired power of the fuel cell stack fan.

[0033] After the control is executed, the actual control effect feedback is obtained. The control effect feedback is the deviation between the coolant outlet temperature measured at the next moment and the temperature at the corresponding position in the predicted temperature field.

[0034] Based on the control effect feedback, the network weights of the multilayer feedforward neural network are adjusted in real time using the online gradient descent backpropagation algorithm to achieve continuous adaptive optimization of the coordinated control of the coolant circulation pump and the fuel cell stack fan.

[0035] As a further aspect of the present invention, based on the control effect feedback, the network weights of the multilayer feedforward neural network are adjusted in real time using an online gradient descent backpropagation algorithm, including:

[0036] Define a loss function, which is the mean square error between the expected coolant outlet temperature and the model-predicted coolant outlet temperature, plus a weighting term for controlling energy consumption.

[0037] In each control cycle, the input variables, the control commands calculated from the current network weights, and the control effect feedback are used together as a training sample.

[0038] The training samples are input into the multilayer feedforward neural network, and the predicted coolant outlet temperature under the current control command is calculated through forward propagation.

[0039] Based on the loss function, calculate the error gradient between the predicted coolant outlet temperature and the actual feedback coolant outlet temperature;

[0040] The error gradient is propagated backward along the multilayer feedforward neural network, and the gradient change of each weight from the output layer, hidden layer to the input layer is calculated sequentially.

[0041] According to a preset learning rate, all network weights of the multilayer feedforward neural network are updated in the opposite direction of the gradient change, thereby realizing the online adaptive mechanism optimization.

[0042] As a further aspect of the present invention, the system further includes:

[0043] A pressure pulsation suppression module is used to perform pressure pulsation suppression processing on the desired rotational speed before the improved fuzzy neural network algorithm generates the coordinated control command for the coolant circulation pump.

[0044] Obtain the historical pressure pulsation spectrum characteristics of the coolant circulation pump at the current speed;

[0045] Based on the historical pressure pulsation spectrum characteristics, predict the frequency point at which the coolant piping system will resonate under the current operating conditions and the desired speed.

[0046] On the reference of the desired rotational speed, a small rotational speed perturbation signal that is incoherent with the frequency point at which resonance occurs is superimposed;

[0047] The superimposed rotational speed is used as the final control speed of the coolant circulation pump, and a corresponding speed control signal is generated.

[0048] As a further aspect of the present invention, the pressure pulsation suppression process for the desired rotational speed further includes performing:

[0049] After the final controlled speed is executed, the actual pressure pulsation signal is collected by a high-frequency pressure sensor in the coolant pipeline;

[0050] Perform a fast Fourier transform on the actual pressure pulsation signal to extract the actual resonant frequency under the current control speed;

[0051] The actual resonant frequency is compared with the predicted frequency point at which resonance occurs, and the frequency prediction deviation is calculated.

[0052] Based on the frequency prediction deviation, the algorithm parameters used to generate the small speed disturbance signal are adaptively adjusted so that the speed disturbance signal superimposed in the next control cycle can more effectively avoid the actual resonant frequency.

[0053] As a further aspect of the present invention, the system further includes:

[0054] The preheating control module is activated to calculate the temperature field based on the improved electrochemical-thermal coupling model during the stack startup phase.

[0055] Based on the temperature field, the water content distribution of the proton exchange membrane is calculated;

[0056] Based on the moisture content distribution, assess the wetting uniformity of each region of the membrane electrode and identify areas at risk of membrane drying.

[0057] Based on the location and degree of dryness of the membrane drying risk area, the initial target setting of the coolant inlet temperature in the improved fuzzy neural network algorithm is adjusted to generate a dedicated fuel cell inlet temperature preheating control curve for the start-up phase.

[0058] Based on the preheating control curve of the fuel cell stack inlet temperature, during the startup phase, the coolant preheater is controlled to gradually increase the temperature of the coolant entering the fuel cell stack until the stack temperature reaches the preset stable operating window.

[0059] As a further aspect of the present invention, the system further includes:

[0060] The steady-state optimization module is used to continuously monitor the electrical impedance spectrum of the membrane electrode during the steady-state operation of the fuel cell stack.

[0061] Based on the changing characteristics of the electrical impedance spectrum, the trend of the wettability of the proton exchange membrane can be determined;

[0062] When the trend of the wet state change indicates that the membrane electrode tends to dry out, the control command for the coolant circulation pump is separated from the cooperative control command generated by the improved fuzzy neural network algorithm.

[0063] Based on the separated control commands for the coolant circulation pump, a coolant flow rate increment compensation based on the changing trend of the humidification state is introduced;

[0064] The compensated coolant circulation pump control command is recombined with the original fuel cell stack fan control command to form a steady-state control command optimized for maintaining membrane humidity.

[0065] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0066] An improved electrochemical-thermal coupling model is built based on an extended equivalent circuit model. This model incorporates all operating condition data, including stack output current, output voltage, coolant temperature and flow, and ambient temperature and humidity, to perform comprehensive calculations of the internal heat generation power distribution and the global temperature field. It overcomes the limitations of traditional simplified models that can only measure the external surface temperature parameters of the stack, deeply linking the internal electrochemical reaction process and heat transfer process of the fuel cell. This fully presents the heat generation variation patterns and temperature distribution characteristics of different regions within the stack, broadening the coverage of stack thermal state analysis, refining the precision of internal thermal parameter analysis, and adapting to the need for accurate characterization of thermal properties under complex and variable operating conditions of fuel cells.

[0067] The fuzzy neural network algorithm incorporates an online adaptive mechanism to dynamically optimize the structure and parameters. It combines the phase change risk region of the coolant within the water-cooled flow channel and the probability of local hotspot generation, synchronously linking the coolant inlet and outlet temperatures with real-time flow parameters to generate coordinated control commands for the coolant circulation pump and the fuel cell stack fan. This changes the traditional operation mode of fixed parameters and rigid control logic. The algorithm can autonomously adjust its computational logic and control weights according to the real-time operating conditions of the fuel cell, achieving synchronous matching and linkage of the operating states of the two types of heat dissipation devices. This adapts to the dynamic adjustment of thermal management throughout the entire fuel cell operation process, ensuring that the control rhythm aligns with the changing patterns of the real-time thermal state inside the fuel cell stack. Attached Figure Description

[0068] Figure 1 This is a timing diagram of the thermal management control system for a water-cooled proton exchange membrane fuel cell described in this invention.

[0069] Figure 2 A flowchart for predicting the risk zone of coolant phase change inside water-cooled flow channels;

[0070] Figure 3 A flowchart illustrating the execution of the improved fuzzy neural network algorithm. Detailed Implementation

[0071] 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.

[0072] 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.

[0073] See Figure 1 This invention provides a thermal management and control system for a water-cooled proton exchange membrane fuel cell, specifically comprising:

[0074] The system comprises a data acquisition module, a model calculation module, a risk prediction module, and a collaborative control module. The data acquisition module collects real-time operating condition data of the fuel cell stack, including stack output current, stack output voltage, coolant inlet temperature, coolant outlet temperature, coolant flow rate, ambient temperature, and ambient humidity. The model calculation module receives this operating condition data and performs calculations using an improved electrochemical-thermal coupling model. This improved electrochemical-thermal coupling model is built based on an extended equivalent circuit model; its input is the operating condition data, and its output is the heat generation power distribution and temperature field inside the stack. The risk prediction module predicts the phase change risk region of the coolant inside the water-cooled flow channel and assesses the probability of local hot spot formation inside the stack based on the heat generation power distribution and temperature field output by the model calculation module. The collaborative control module employs an improved fuzzy neural network algorithm; its inputs are the phase change risk region and local hot spot formation probability output by the risk prediction module, as well as the coolant inlet temperature, coolant outlet temperature, and coolant flow rate provided by the data acquisition module. The improved fuzzy neural network algorithm is optimized based on an online adaptive mechanism. By processing the input information, it generates coordinated control commands for the coolant circulation pump speed and the fuel cell stack fan power, thereby achieving dynamic and optimized control of the thermal management system.

[0075] In one embodiment of the present invention, the real-time ohmic internal resistance, activation internal resistance, and concentration internal resistance of the fuel cell stack are calculated using the stack output current and stack output voltage, and an extended equivalent circuit model is established, which includes the mapping relationship between the three internal resistance parameters and the current density. Based on this extended equivalent circuit model, the energy loss generated by activation polarization, ohmic polarization, and concentration polarization during stack operation is calculated, and this energy loss is converted into the real-time heat source intensity of each region of the stack. A three-dimensional computational grid is divided inside the stack along the gas flow channel direction and the membrane electrode stack direction. A three-dimensional heat transfer network model is established, including the solid structure, membrane electrode, gas diffusion layer, bipolar plate, and cooling channel. The real-time heat source intensity is used as the internal heat source and loaded onto the corresponding grid position in the three-dimensional heat transfer network model. The coolant inlet temperature, coolant outlet temperature, coolant flow rate, ambient temperature, and ambient humidity are used as the boundary conditions of the three-dimensional heat transfer network model. Solving the three-dimensional heat transfer network model yields the temperature distribution of each solid component and fluid region inside the stack, which is the heat generation power distribution and temperature field inside the stack.

[0076] In one specific implementation, the model calculation module receives real-time operating condition data from the data acquisition module. This data includes the fuel cell stack output current, output voltage, coolant inlet temperature, coolant outlet temperature, coolant flow rate, ambient temperature, and ambient humidity. Using the fuel cell stack output current and voltage, the model calculation module calculates the fuel cell stack's real-time ohmic internal resistance, real-time activation internal resistance, and real-time concentration internal resistance based on the equivalent circuit principle and a real-time online parameter identification method. In this specific implementation, an extended equivalent circuit model is established that includes the mapping relationship between the above three internal resistance parameters and current density. This extended equivalent circuit model can be expressed as:

[0077]

[0078] in: It is the output voltage of the fuel cell stack. It is the open-circuit voltage of the fuel cell stack. It is the output current of the fuel cell stack. It is the real-time ohmic internal resistance that varies with current density. The activation overpotential is determined by the real-time activation internal resistance. The concentration overpotential is determined by the real-time concentration internal resistance. The extended equivalent circuit model, by establishing a dynamic mapping relationship between the internal resistance parameter and the current density, can more accurately describe the internal state of the fuel cell stack under different loads.

[0079] Based on the extended equivalent circuit model, the model calculation module calculates the energy loss caused by activation polarization, ohmic polarization, and concentration polarization during the operation of the fuel cell stack. This energy loss is the product of the voltage drop caused by polarization and the output current. In specific implementation, the model calculation module converts this energy loss into the real-time heat source intensity of each region of the fuel cell stack. During the conversion process, the volume proportion and heat generation contribution coefficient of each component inside the fuel cell stack, including the membrane electrode assembly (MEA) and bipolar plates, are considered, and the total heat generation power is distributed according to spatial location. The model calculation module divides a three-dimensional computational mesh inside the fuel cell stack along the gas flow channel direction and the MEA stack direction. The size of the three-dimensional computational mesh is set according to the fuel cell stack geometry and computational accuracy requirements. The model calculation module establishes a three-dimensional heat transfer network model including the solid structure, MEA, gas diffusion layer, bipolar plates, and cooling channels. The three-dimensional heat transfer network model discretizes the fuel cell stack into multiple interconnected nodes, and the heat conduction, heat convection, and heat radiation relationships between nodes are modeled through a thermal resistance network.

[0080] The model calculation module uses the real-time heat source intensity as the internal heat source and loads it onto the corresponding grid position in the three-dimensional heat transfer network model. The loading position corresponds to the membrane electrode reactive region. In specific implementation, the model calculation module uses the coolant inlet temperature, coolant outlet temperature, coolant flow rate, ambient temperature, and ambient humidity as boundary conditions for the three-dimensional heat transfer network model. The coolant inlet temperature and coolant flow rate are used to set the convective heat transfer conditions at the cooling channel inlet, while the ambient temperature and ambient humidity are used to set the natural convection and radiation heat transfer boundary conditions between the stack's outer wall and the air. The model calculation module uses numerical iterative methods, such as the finite volume method or the finite element method, to solve the three-dimensional heat transfer network model. The process of solving the three-dimensional heat transfer network model involves solving a set of algebraic equations describing the energy balance of each node, including internal heat source terms and all boundary conditions. After solving, the model calculation module obtains the temperature distribution of each solid component and fluid region inside the stack. This temperature distribution is a visualization and quantitative expression of the heat generation power distribution and temperature field inside the stack, providing input for the subsequent risk prediction module.

[0081] In one embodiment of the present invention, see [reference] Figure 2The temperature distribution in the cooling channel region of the 3D heat transfer network model is extracted, and the temperature values ​​of the inner wall of the cooling channel and their gradient along the flow direction are identified. The working pressure of the coolant at the inlet of the cooling channel is obtained, and the saturation temperature of the coolant at that working pressure is queried based on the inlet temperature. The temperature value of the inner wall of the cooling channel is compared with the saturation temperature point by point. If the wall temperature is higher than the saturation temperature, the corresponding location is identified as a potential phase change risk point. For all identified potential phase change risk points, the temperature change rate along the channel axis is calculated. If the temperature change rate exceeds a threshold, the potential phase change risk point is identified as having the risk of local boiling leading to bubble accumulation and is marked as a phase change risk region. The temperature distribution in the membrane electrode, catalyst layer, and proton exchange membrane regions of the 3D heat transfer network model is extracted, and the local temperature peaks and their spatial coordinates in the membrane electrode, catalyst layer, and proton exchange membrane regions are obtained. The difference between the local temperature peak and the volume-weighted average temperature of the entire stack is calculated to obtain the local superheat. The risk level of local temperature runaway is assessed based on the local overheating and the temperature distribution uniformity index of the membrane electrode, catalyst layer, and proton exchange membrane regions. Based on this risk level and the current stack output current fluctuation trend, the quantified probability of irreversible high-temperature damage to the membrane electrode in the next control cycle is calculated; this quantified probability is the probability of local hotspot formation.

[0082] The risk prediction module receives the heat generation power distribution and temperature field inside the fuel cell stack from the model calculation module. In specific implementations, the risk prediction module extracts the temperature distribution of the cooling channel region in the three-dimensional heat transfer network model. The temperature distribution exists in the form of an array of temperature values ​​for grid nodes. The risk prediction module identifies the temperature values ​​of the inner wall surface of the cooling channel and their gradient along the flow direction. Specifically, it obtains the temperature of the solid grid wall surface nodes in contact with the coolant and calculates the temperature difference between adjacent nodes along the flow direction to obtain the gradient. The module obtains the working pressure of the coolant at the inlet of the cooling channel. The working pressure is directly measured by a pressure sensor installed in the cooling pipeline. Based on the coolant inlet temperature, the module queries the coolant property database to obtain the saturation temperature of the coolant at the working pressure. In some embodiments, the risk prediction module compares the temperature values ​​of the inner wall surface of the cooling channel with the saturation temperature point by point. If the temperature value of a certain wall surface node is higher than the queried saturation temperature, the location represented by that node is determined to be a potential phase change risk point. For all identified potential phase change risk points, the risk prediction module calculates the rate of temperature change along the flow channel axis. This rate of temperature change is obtained by calculating the ratio of the temperature difference between the point and its upstream neighbor to the distance. It can be understood that if the calculated rate of temperature change exceeds a preset threshold, this potential phase change risk point is determined to have a risk of bubble accumulation due to localized boiling. The risk prediction module then marks this point and its adjacent area as a phase change risk region, and the spatial coordinate information of the phase change risk region is output.

[0083] The risk prediction module extracts the temperature distribution of the three-dimensional heat transfer network model in the membrane electrode, catalyst layer, and proton exchange membrane regions, obtaining the local temperature peaks and their spatial coordinates in these regions. The difference between the local temperature peaks and the volume-weighted average temperature of the entire fuel cell stack is calculated to obtain the local superheat.

[0084]

[0085] in: Represents the local temperature peak. This represents the volume-weighted average temperature of the entire fuel cell stack. In some embodiments, the risk level of local temperature runaway is assessed based on local overheating and the temperature distribution uniformity index of the membrane electrode, catalyst layer, and proton exchange membrane regions. The temperature distribution uniformity index is calculated by statistically analyzing the standard deviation of the temperatures of all nodes within a specific region. The risk prediction module calculates the quantified probability of irreversible high-temperature damage to the membrane electrode in the next control cycle based on the risk level of local temperature runaway and the fluctuation trend of the current fuel cell stack output current. Optionally, one way to calculate the quantified probability is to map the risk level to a base probability value, and then weight and correct this base probability value according to the rising slope of the fuel cell stack output current over a recent period. The corrected probability value is the probability of local hotspot generation. It can be understood that the probability of local hotspot generation is output as a value between 0 and 1 for use by the cooperative control module.

[0086] In one embodiment of the present invention, see [reference] Figure 3The input variables for the input layer are the area ratio of the phase transition risk region, the probability of local hotspot generation, the coolant inlet temperature, the coolant outlet temperature, the coolant flow rate, the ambient temperature, and the ambient humidity. The input variables undergo fuzzification preprocessing, transforming each input variable into a membership value of multiple fuzzy subsets through multiple fuzzy membership functions. These membership values ​​are then input into a multilayer feedforward neural network containing at least one hidden layer, with its weights initially set based on expert experience. The output of the multilayer feedforward neural network is the desired speed of the coolant circulation pump and the desired power of the fuel cell stack fan. After control is executed, the actual control effect feedback is obtained, which is the deviation between the measured coolant outlet temperature and the corresponding temperature in the predicted temperature field at the next moment. Based on the control effect feedback, the network weights of the multilayer feedforward neural network are adjusted in real time using an online gradient descent backpropagation algorithm to achieve continuous adaptive optimization of the coordinated control of the coolant circulation pump and the fuel cell stack fan. A loss function is defined as the mean square error between the desired coolant outlet temperature and the model-predicted coolant outlet temperature, plus a weighted term for control energy consumption. In each control cycle, the input variables, the control command calculated from the current network weights, and the control effect feedback are used as training samples. These training samples are input into a multi-layer feedforward neural network, and forward propagation calculates the predicted coolant outlet temperature under the current control command. Based on the loss function, the error gradient between the predicted coolant outlet temperature and the actual feedback coolant outlet temperature is calculated. This error gradient is then propagated backward along the multi-layer feedforward neural network, calculating the gradient changes of each weight from the output layer, hidden layer, to the input layer. Following a preset learning rate, all network weights of the multi-layer feedforward neural network are updated in the opposite direction of the gradient changes, achieving online adaptive optimization.

[0087] The collaborative control module receives the area proportion of the phase transition risk region and the probability of local hotspot generation from the risk prediction module, and simultaneously receives the coolant inlet temperature, coolant outlet temperature, coolant flow rate, ambient temperature, and ambient humidity from the data acquisition module. In specific implementations, the collaborative control module uses the area proportion of the phase transition risk region, the probability of local hotspot generation, the coolant inlet temperature, the coolant outlet temperature, the coolant flow rate, the ambient temperature, and the ambient humidity as seven input variables for the input layer. The collaborative control module performs fuzzification preprocessing on each input variable, transforming each input variable into membership values ​​of multiple fuzzy subsets through multiple fuzzy membership functions. In some embodiments, the input variable "coolant outlet temperature" can be divided into three fuzzy subsets: "low," "medium," and "high," and its membership value belonging to each subset is calculated using a triangular or Gaussian membership function. See Table 1 for input variables and their typical fuzzy subset divisions.

[0088] Table 1: Fuzzy Subset Partition Table for Input Variables

[0089] Input variables Fuzzy subset 1 Fuzzy subset 2 Fuzzy subset 3 Percentage of area at risk of phase transition Small middle big Local hotspot generation probability Low middle high Coolant inlet temperature Low suitable High Coolant outlet temperature Low middle high Coolant flow rate Small middle big Ambient temperature Low middle high Ambient humidity dry Moderate damp

[0090] The collaborative control module inputs all membership values ​​obtained after fuzzification preprocessing into a multilayer feedforward neural network. In specific implementations, the multilayer feedforward neural network contains at least one hidden layer, the number of neurons in which can be configured according to the system complexity, and the network weights of the multilayer feedforward neural network are set in the initial stage based on expert experience or historical data training. The output layer of the multilayer feedforward neural network produces two explicit output values: the desired speed of the coolant circulation pump and the desired power of the fuel cell stack fan. After the collaborative control module executes the control command consisting of the desired speed and desired power, it obtains the actual control effect feedback from the data acquisition module. Specifically, the control effect feedback is the deviation between the actual value of the coolant outlet temperature measured at the next sampling time and the temperature at the corresponding location in the temperature field predicted by the improved electrochemical-thermal coupling model.

[0091] The collaborative control module, based on control effect feedback, utilizes an online gradient descent backpropagation algorithm to adjust the network weights of a multi-layer feedforward neural network in real time. The collaborative control module defines a loss function to guide the adjustment of the network weights; the expression for the loss function is as follows:

[0092]

[0093] in: This represents the coolant outlet temperature predicted by the model. This represents the actual coolant outlet temperature. The temperature reference value is used for normalization. This represents the power consumption of the coolant circulation pump. This represents the power consumption of the fuel cell stack fan. The power reference value is used for normalization. This refers to the dimensionless coefficients controlling the relative weights of the two components. In each control cycle, the collaborative control module uses the current input variable, the control command calculated from the current network weights, and the acquired control effect feedback as a training sample. The collaborative control module then inputs the training sample into a multi-layer feedforward neural network to perform forward propagation calculations, obtaining the coolant outlet temperature predicted by the neural network under the current control command. It is understandable that the collaborative control module, based on the loss function... The error gradient between the predicted coolant outlet temperature and the actual feedback coolant outlet temperature is calculated. The collaborative control module backpropagates the error gradient along a multi-layer feedforward neural network, sequentially calculating the gradient change of each weight from the output layer, hidden layer to the input layer. ,in, Represents the connection weights of any layer in the neural network. The collaborative control module operates according to a preset learning rate. Update all network weights of the multilayer feedforward neural network in the opposite direction of the gradient change, according to the following update rule: ,in, The network weights before the update. This represents the updated network weights. The collaborative control module optimizes the network through continuous weight updates, enabling an online adaptive mechanism.

[0094] In one embodiment of the present invention, pressure pulsation suppression processing is performed on the desired speed before the improved fuzzy neural network algorithm generates the coordinated control command for the coolant circulation pump. Historical pressure pulsation spectrum characteristics at the current speed of the coolant circulation pump are obtained. Based on the historical pressure pulsation spectrum characteristics, the frequency point at which the coolant piping system will resonate under the current operating conditions and the desired speed is predicted. A small speed disturbance signal, incoherent with the resonant frequency point, is superimposed on the reference of the desired speed. The superimposed speed is used as the final control speed of the coolant circulation pump, and a corresponding speed control signal is generated. After executing the final control speed, the actual pressure pulsation signal is acquired through a high-frequency pressure sensor in the coolant piping. A fast Fourier transform is performed on the actual pressure pulsation signal to extract the actual resonant frequency at the current control speed. The actual resonant frequency is compared with the predicted resonant frequency point, and the frequency prediction deviation is calculated. Based on the frequency prediction deviation, the algorithm parameters used to generate the small speed disturbance signal are adaptively adjusted so that the speed disturbance signal superimposed in the next control cycle can more effectively avoid the actual resonant frequency.

[0095] Before the improved fuzzy neural network algorithm generates the coordinated control commands for the coolant circulation pump, the pressure pulsation suppression module performs pressure pulsation suppression processing on the desired speed of the coolant circulation pump. In specific implementations, the pressure pulsation suppression module acquires the historical pressure pulsation spectrum characteristics of the coolant circulation pump at the current speed. These historical pressure pulsation spectrum characteristics are extracted from time-series data recorded by a high-frequency pressure sensor through historical data analysis, identifying the dominant frequency and amplitude characteristics of pressure oscillations in the coolant piping system under different speed conditions. In some embodiments, the historical pressure pulsation spectrum characteristics can be stored and retrieved in the form of a data table, as shown in Table 2.

[0096] Table 2: Historical Pressure Pulsation Spectral Characteristics

[0097] Coolant circulation pump speed (rpm) Dominant frequency 1 (Hz) Amplitude 1 (kPa) Dominant frequency 2 (Hz) Amplitude 2 (kPa) 2000 85.5 1.2 167.3 0.8 2500 102.1 1.5 199.8 1.1 3000 125.7 2.0 245.0 1.3

[0098] The pressure pulsation suppression module predicts the frequency points at which the coolant piping system may resonate under current operating conditions and the desired speed of the coolant circulation pump, based on historical pressure pulsation spectrum characteristics. The module then superimposes a small speed disturbance signal, independent of the predicted resonant frequency points, onto the desired speed of the coolant circulation pump. This small speed disturbance signal is a sinusoidal or random signal with adjustable amplitude and frequency. Essentially, the pressure pulsation suppression module uses the speed after superimposing the small speed disturbance signal as the final control speed of the coolant circulation pump and generates a corresponding speed control signal, which is output to the coolant circulation pump driver. Optionally, the process of superimposing the small speed disturbance signal can be described by the following relationship:

[0099]

[0100] in: Represents the final controlled speed. This represents the desired rotational speed output by the collaborative control module. The amplitude of the signal representing a small speed disturbance is... The frequency representing a minute speed disturbance signal. Represents time, This represents the initial phase of a small speed disturbance signal.

[0101] After the final controlled speed is achieved, the pressure pulsation suppression module acquires the actual pressure pulsation signal through a high-frequency pressure sensor in the coolant piping. The module performs a Fast Fourier Transform on the actual pressure pulsation signal to extract the actual resonant frequency of the coolant piping system at the current controlled speed. In some embodiments, the pressure pulsation suppression module compares the extracted actual resonant frequency with the previously predicted frequency point that would generate resonance, calculating the frequency prediction deviation. ,in, This represents the predicted frequency point where resonance occurs. This represents the actual resonant frequency. The pressure pulsation suppression module predicts the deviation based on the calculated frequency. The algorithm parameters used to generate small speed disturbance signals are adaptively adjusted. It can be understood that the goal of adjusting the algorithm parameters is to ensure that the frequency of the speed disturbance signal superimposed in the next control cycle more effectively avoids the latest measured actual resonant frequency. This allows for continuous and optimized suppression of pressure pulsations.

[0102] In one embodiment of the present invention, during the fuel cell stack startup phase, a temperature field is calculated based on an improved electrochemical-thermal coupling model. Based on the temperature field, the water content distribution of the proton exchange membrane is calculated. According to the water content distribution, the wetting uniformity of each region of the membrane electrode is assessed, and membrane drying risk areas are identified. Based on the location and degree of dryness of the membrane drying risk areas, the initial target setting for the coolant inlet temperature in the improved fuzzy neural network algorithm is adjusted to generate a dedicated fuel cell stack inlet temperature preheating control curve for the startup phase. According to the fuel cell stack inlet temperature preheating control curve, during the startup phase, the coolant preheater is controlled to perform a gradient temperature increase on the coolant entering the fuel cell until the fuel cell temperature reaches a preset stable operating window. During the fuel cell stack steady-state operation phase, the electrical impedance spectrum of the membrane electrode is continuously monitored. Based on the changing characteristics of the electrical impedance spectrum, the trend of the proton exchange membrane's wetting state is determined. When the trend of the wetting state indicates that the membrane electrode has a tendency to dry out, control commands for the coolant circulation pump are separated from the collaborative control commands generated by the improved fuzzy neural network algorithm. Based on the separated control commands for the coolant circulation pump, a coolant flow rate increment compensation based on the trend of humidification changes is introduced. The compensated coolant circulation pump control commands are then recombined with the original stack fan control commands to form steady-state control commands optimized for membrane humidity maintenance.

[0103] In specific implementation, the preheating control module is activated during the stack startup phase. Based on the temperature field calculated by the improved electrochemical-thermal coupling model, the preheating control module performs subsequent operations. Based on the temperature field, it calculates the water content distribution of the proton exchange membrane. The calculation process is based on the water transfer and phase transition equilibrium equation of the proton exchange membrane, combined with the temperature values ​​of the membrane electrode region in the temperature field. In some embodiments, the preheating control module assesses the wetting uniformity of each region of the membrane electrode based on the calculated water content distribution, identifying membrane drying risk areas. The assessment method can be to compare the water content of each calculation unit with a preset critical water content threshold; areas below this threshold are identified as membrane drying risk areas. Based on the location and degree of dryness of the membrane drying risk areas, the preheating control module adjusts the initial target setting of the coolant inlet temperature in the improved fuzzy neural network algorithm. Specifically, the adjustment strategy is to increase the set value of the coolant inlet temperature to enhance preheating based on the location and degree of dryness of the driest region. Based on the adjusted target settings, the startup preheating control module generates a dedicated stack inlet temperature preheating control curve for the startup phase. This curve represents a gradual increase in temperature over time. Following this curve, the startup preheating control module controls the coolant preheater to gradually increase the temperature of the coolant entering the stack during the startup phase, until the stack temperature reaches the preset stable operating window.

[0104] The steady-state optimization module operates continuously during the steady-state operation of the fuel cell stack. In practice, it continuously monitors the electrical impedance spectrum of the membrane electrode assembly (MEA) via a fixed-frequency scanning method using an electrochemical workstation mounted on the stack. Based on the changing characteristics of the electrical impedance spectrum, the module determines the wetting trend of the proton exchange membrane. This determination can be made by analyzing changes in characteristic frequency points in the electrical impedance spectrum or changes in the fitting parameters of the equivalent circuit model to infer whether the membrane humidity is trending towards wetting or drying. When the wetting trend indicates a tendency for the MEA to dry out, the module extracts control commands for the coolant circulation pump from the cooperative control commands generated by the improved fuzzy neural network algorithm. Based on these extracted control commands, the module introduces a coolant flow rate increment compensation based on the wetting trend. It can be understood that the calculation of coolant flow rate increment compensation can be expressed as:

[0105]

[0106] in: It is the compensation coefficient. This is the relative humidity reference value corresponding to the target membrane humidity. The current membrane humidity characterization value is derived from the electrical impedance spectrum. The steady-state optimization module recombines the compensated coolant circulation pump control command with the original stack fan control command output by the improved fuzzy neural network algorithm to form a steady-state control command optimized for membrane humidity maintenance. This command is ultimately output to the actuator.

[0107] 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 thermal management control system for a water-cooled proton exchange membrane fuel cell, characterized in that, include: The data acquisition module acquires the operating condition data of the fuel cell stack, including the stack output current, stack output voltage, coolant inlet temperature, coolant outlet temperature, coolant flow rate, ambient temperature, and ambient humidity. The model calculation module calculates the heat generation power distribution and temperature field inside the fuel cell stack based on the operating condition data using an improved electrochemical-thermal coupling model. The improved electrochemical-thermal coupling model is constructed based on an extended equivalent circuit model. The risk prediction module predicts the phase change risk area of ​​the coolant inside the water-cooled channel based on the heat generation power distribution and temperature field inside the fuel cell stack, and assesses the probability of local hot spot generation. The collaborative control module employs an improved fuzzy neural network algorithm to generate collaborative control commands for the coolant circulation pump and the fuel cell stack fan based on the phase transition risk region, the probability of local hotspot generation, the coolant inlet temperature, the coolant outlet temperature, and the coolant flow rate. The improved fuzzy neural network algorithm is optimized based on an online adaptive mechanism.

2. The thermal management control system for a water-cooled proton exchange membrane fuel cell according to claim 1, characterized in that, The improved electrochemical-thermal coupling model includes the following steps: Using the output current and output voltage of the fuel cell stack, the real-time ohmic internal resistance, activation internal resistance, and concentration internal resistance of the fuel cell stack are calculated, and an extended equivalent circuit model containing the mapping relationship between the three internal resistance parameters and the current density is established. Based on the extended equivalent circuit model, the energy loss caused by activation polarization, ohmic polarization and concentration polarization during the operation of the fuel cell stack is calculated, and this energy loss is converted into the real-time heat source intensity of each region of the fuel cell stack. A three-dimensional computational mesh was generated inside the fuel cell stack along the gas flow path and the membrane electrode stack direction. A three-dimensional heat transfer network model was established, including a solid structure, a membrane electrode, a gas diffusion layer, a bipolar plate, and cooling channels. The intensity of the real-time heat source is used as an internal heat source and loaded into the corresponding grid position in the three-dimensional heat transfer network model; The coolant inlet temperature, coolant outlet temperature, coolant flow rate, ambient temperature, and ambient humidity are used as boundary conditions for the three-dimensional heat transfer network model. Solving the three-dimensional heat transfer network model yields the temperature distribution of each solid component and fluid region inside the fuel cell stack, i.e., the heat generation power distribution and temperature field inside the fuel cell stack.

3. The thermal management control system for a water-cooled proton exchange membrane fuel cell according to claim 2, characterized in that, The predicted phase change risk area for the coolant inside the water-cooled flow channel includes: Extract the temperature distribution of the cooling channel region in the three-dimensional heat transfer network model, and identify the temperature value of the inner wall of the cooling channel and its gradient along the flow direction; Obtain the working pressure of the coolant at the inlet of the cooling channel, and based on the coolant inlet temperature, query the saturation temperature of the coolant at the working pressure; The temperature value of the inner wall of the cooling channel is compared with the saturation temperature point by point. If the wall temperature is higher than the saturation temperature, the corresponding location is determined to be a potential phase change risk point. For all identified potential phase change risk points, calculate their temperature change rate along the flow channel axis. If the temperature change rate exceeds the threshold, the potential phase change risk point is determined to have the risk of local boiling leading to bubble accumulation, and is marked as the phase change risk area.

4. The thermal management control system for a water-cooled proton exchange membrane fuel cell according to claim 3, characterized in that, The assessment of the probability of local hotspot generation includes: The temperature distribution of the three-dimensional heat transfer network model in the membrane electrode, catalyst layer and proton exchange membrane regions is extracted to obtain the local temperature peaks and their spatial coordinates in the membrane electrode, catalyst layer and proton exchange membrane regions; The local superheat is obtained by calculating the difference between the local temperature peak and the volume-weighted average temperature of the entire fuel cell stack. The risk level of local temperature runaway is assessed based on the local overheating and the temperature distribution uniformity index of the membrane electrode, catalyst layer, and proton exchange membrane regions. Based on the risk level of the local temperature runaway and combined with the fluctuation trend of the current stack output current, the quantitative probability of irreversible high-temperature damage to the membrane electrode in the next control cycle is calculated, that is, the probability of local hot spot generation.

5. The thermal management control system for a water-cooled proton exchange membrane fuel cell according to claim 1, characterized in that, The improved fuzzy neural network algorithm includes the following steps: The area ratio of the phase transition risk region, the probability of local hotspot generation, the coolant inlet temperature, the coolant outlet temperature, the coolant flow rate, the ambient temperature, and the ambient humidity are used as input variables for the input layer. The input variables are preprocessed by fuzzification, and each input variable is transformed into the membership value of multiple fuzzy subsets through multiple fuzzy membership functions; The membership value is input into a multilayer feedforward neural network, which contains at least one hidden layer, and the network weights are set according to expert experience in the initial stage. The output of the multilayer feedforward neural network is the desired speed of the coolant circulation pump and the desired power of the fuel cell stack fan. After the control is executed, the actual control effect feedback is obtained. The control effect feedback is the deviation between the coolant outlet temperature measured at the next moment and the temperature at the corresponding position in the predicted temperature field. Based on the control effect feedback, the network weights of the multilayer feedforward neural network are adjusted in real time using the online gradient descent backpropagation algorithm to achieve continuous adaptive optimization of the coordinated control of the coolant circulation pump and the fuel cell stack fan.

6. The thermal management control system for a water-cooled proton exchange membrane fuel cell according to claim 5, characterized in that, Based on the control effect feedback, the network weights of the multilayer feedforward neural network are adjusted in real time using the online gradient descent backpropagation algorithm, including: Define a loss function, which is the mean square error between the expected coolant outlet temperature and the model-predicted coolant outlet temperature, plus a weighting term for controlling energy consumption. In each control cycle, the input variables, the control commands calculated from the current network weights, and the control effect feedback are used together as a training sample. The training samples are input into the multilayer feedforward neural network, and the predicted coolant outlet temperature under the current control command is calculated through forward propagation. Based on the loss function, calculate the error gradient between the predicted coolant outlet temperature and the actual feedback coolant outlet temperature; The error gradient is propagated backward along the multilayer feedforward neural network, and the gradient change of each weight from the output layer, hidden layer to the input layer is calculated sequentially. According to a preset learning rate, all network weights of the multilayer feedforward neural network are updated in the opposite direction of the gradient change, thereby realizing the online adaptive mechanism optimization.

7. The thermal management control system for a water-cooled proton exchange membrane fuel cell according to claim 5, characterized in that, The system also includes: A pressure pulsation suppression module is used to perform pressure pulsation suppression processing on the desired rotational speed before the improved fuzzy neural network algorithm generates the coordinated control command for the coolant circulation pump. Obtain the historical pressure pulsation spectrum characteristics of the coolant circulation pump at the current speed; Based on the historical pressure pulsation spectrum characteristics, predict the frequency point at which the coolant piping system will resonate under the current operating conditions and the desired rotational speed. On the reference of the desired rotational speed, a small rotational speed perturbation signal that is incoherent with the frequency point at which resonance occurs is superimposed; The superimposed rotational speed is used as the final control speed of the coolant circulation pump, and a corresponding speed control signal is generated.

8. The thermal management control system for a water-cooled proton exchange membrane fuel cell according to claim 7, characterized in that, The pressure pulsation suppression process for the desired rotational speed also includes performing: After the final controlled speed is executed, the actual pressure pulsation signal is collected by a high-frequency pressure sensor in the coolant pipeline; Perform a fast Fourier transform on the actual pressure pulsation signal to extract the actual resonant frequency under the current control speed; The actual resonant frequency is compared with the predicted frequency point at which resonance occurs, and the frequency prediction deviation is calculated. Based on the frequency prediction deviation, the algorithm parameters used to generate the small speed disturbance signal are adaptively adjusted so that the speed disturbance signal superimposed in the next control cycle can more effectively avoid the actual resonant frequency.

9. The thermal management control system for a water-cooled proton exchange membrane fuel cell according to claim 1, characterized in that, The system also includes: The preheating control module is activated to calculate the temperature field based on the improved electrochemical-thermal coupling model during the stack startup phase. Based on the temperature field, the water content distribution of the proton exchange membrane is calculated; Based on the moisture content distribution, assess the wetting uniformity of each region of the membrane electrode and identify areas at risk of membrane drying. Based on the location and degree of dryness of the membrane drying risk area, the initial target setting of the coolant inlet temperature in the improved fuzzy neural network algorithm is adjusted to generate a dedicated fuel cell inlet temperature preheating control curve for the start-up phase. Based on the preheating control curve of the fuel cell stack inlet temperature, during the startup phase, the coolant preheater is controlled to gradually increase the temperature of the coolant entering the fuel cell stack until the stack temperature reaches the preset stable operating window.

10. The thermal management control system for a water-cooled proton exchange membrane fuel cell according to claim 9, characterized in that, The system also includes: The steady-state optimization module is used to continuously monitor the electrical impedance spectrum of the membrane electrode during the steady-state operation of the fuel cell stack. Based on the changing characteristics of the electrical impedance spectrum, the trend of the wettability of the proton exchange membrane can be determined; When the trend of the wet state change indicates that the membrane electrode tends to dry out, the control command for the coolant circulation pump is separated from the cooperative control command generated by the improved fuzzy neural network algorithm. Based on the separated control commands for the coolant circulation pump, a coolant flow rate increment compensation based on the changing trend of the humidification state is introduced; The compensated coolant circulation pump control command is recombined with the original stack fan control command to form a steady-state control command optimized for membrane humidity maintenance.

Images