A method for diagnosing water management faults in a vehicle fuel cell power system

Abstract

The present invention provides a method for diagnosing water management faults in a fuel cell power system for a vehicle, belonging to the field of fuel cell vehicle technology. The method first screens a data sample set of normal operating conditions from the system's historical operating data, selecting operating parameters to form a training data set. The training data set is then subjected to outlier processing, noise processing, and data normalization, selecting parameters as the input and expected output of a FNN voltage model and a RNN anode voltage drop model, respectively, for training. Based on the residual between the output of the data-driven model and the actual output during system operation, a water management fault classifier is constructed to diagnose water management faults. The present invention utilizes historical system operating data to train a data-driven model for water management fault diagnosis in the power system, significantly reducing modeling difficulty and improving model accuracy and versatility. Water management fault classification logic is then constructed based on empirical knowledge, enabling rapid diagnosis of normal operation, flooding faults, and drying faults.

Description

Technical Field

[0001] The present invention belongs to the technical field of fuel cell vehicles, and in particular relates to a method for diagnosing water management faults in a vehicle fuel cell power system. Background Art

[0002] The proton exchange membrane fuel cell (PEMFC) is an electrochemical energy conversion device that converts chemical energy into electrical energy. It offers advantages such as environmental friendliness, high energy density, rapid start-up at low temperatures, and short hydrogen refueling times. It holds broad application prospects as a power source for medium and large vehicle power systems. Currently, the stability and durability of PEMFCs are the main bottlenecks for the widespread adoption of automotive fuel cell power systems. Water management failures are the most common type of failure during system operation. In the short term, they can affect the stability of the system's output power, and over time, they can cause irreversible damage to the battery.

[0003] Due to the complex multi-mechanism water transport characteristics of proton exchange membranes and the complex dynamic phase transitions of water within the cell, water management fault diagnosis has always been a major challenge in fuel cell research. Accurate and reliable water management fault diagnosis provides guidance for developing troubleshooting strategies. Therefore, finding a highly accurate, equipment-free water management fault diagnosis method is crucial for improving system reliability and durability.

[0004] The existing mainstream fuel cell water management fault diagnosis technology is based on the mechanism model. The system block diagram of this method is as follows: Figure 1 As shown in the figure, the main problems are that the mechanism model is difficult to model, has low versatility, requires a lot of calculations, and some parameters in the model, such as hydrogen and oxygen partial pressures, are difficult to measure in real time in existing systems. Summary of the Invention

[0005] In response to the above problems, the present invention proposes a method for diagnosing water management faults in automotive fuel cell power systems. By utilizing the system's historical operating data and the powerful nonlinear mapping function of artificial neural networks, a data-driven model is used instead of a mechanism model to achieve water management fault diagnosis. The method has the significant advantages of simple modeling, strong versatility, low computational complexity, suitability for real-time processing in embedded systems, and no additional equipment required.

[0006] The specific technical solutions of the present invention are as follows:

[0007] A method for diagnosing water management faults in a vehicle fuel cell power system, comprising the following steps:

[0008] Step 1: Filter the data sample set in normal operation state from the system historical operation data;

[0009] Step 2: Select operating parameters from the data sample set to form a training data set; wherein the operating parameters include current, cooling water inlet temperature, cooling water outlet temperature, anode inlet pressure, anode outlet pressure, hydrogen circulation flow, air flow, and node voltage;

[0010] Step 3: Use the absolute median deviation method based on sliding window to deal with outliers in the training data set;

[0011] Step 4: Use the moving average smoothing filter method to perform noise processing on the training data set processed in step 3;

[0012] Step 5: Use the minimum-maximum normalization method to normalize the training data set processed in step 4;

[0013] Step 6: Selecting the first part of the operating parameters in the training data set processed in step 5 to form a voltage model input data set; wherein the first part of the operating parameters includes current, cooling water inlet temperature, cooling water outlet temperature, anode inlet pressure, hydrogen circulation flow rate, and air flow rate;

[0014] Step 7: Select the node voltage in the training data set processed in step 5 as the voltage model output data set;

[0015] Step 8: Select the second part of parameters in the training data set processed in step 5 to form the anode pressure drop model input data set; wherein the second part of operating parameters includes current, cooling water inlet temperature, cooling water outlet temperature, anode inlet pressure, and hydrogen circulation flow rate;

[0016] Step 9: Select the difference between the anode inlet pressure and the anode outlet pressure in the training data set processed in step 5 as the output data set of the anode pressure drop model;

[0017] Step 10: Using the voltage model input dataset described in step 6 as input and the voltage model output dataset described in step 7 as the desired output, a shallow fully connected feedforward neural network (FNN) is trained based on the minimum gradient descent method to obtain an FNN voltage model; wherein the training loss function used is the mean Euclidean distance between the FNN output and the desired output;

[0018] Step 11: Using the anode pressure drop model input dataset described in step 8 as input and the anode pressure drop model output dataset described in step 9 as the desired output, a shallow fully connected recurrent neural network (RNN) is trained based on the minimum gradient descent method to obtain the RNN anode pressure drop model; wherein the training loss function used is the mean Euclidean distance between the RNN output and the desired output;

[0019] Step 12: The FNN voltage model and the RNN anode voltage drop model are used in parallel to form a data-driven model. Based on the residual between the output of the data-driven model and the actual output of the system operation process, a water management fault classifier is constructed; wherein the output of the data-driven model includes the expected output voltage Expected node voltage standard deviation and the expected anode pressure drop The actual output of the system operation process includes the actual output voltage U, the actual voltage standard deviation U STD and actual anode pressure drop ΔP AN ;

[0020] Water management fault diagnosis is completed based on the water management fault classifier. The specific steps are as follows:

[0021] Step 12.1: Preset and Corresponding to U, U STD and ΔP AN The upper limit of normality of the residual i ;where, i={U,U STD ,ΔP AN};

[0022] Step 12.2: Calculate separately and Corresponding to U and U at the current moment STD and ΔP AN The residual in,

[0023] Step 12.3: Determine ε U >l U and Is it true? If so, go to step 12.4; if false, output the diagnosis result as normal;

[0024] Step 12.4: Judgment Is it true? If it is true, the output diagnosis result is flooding; if it is false, the output diagnosis result is dry.

[0025] Furthermore, the specific criteria for screening in step 1 are: setting the time window T, the lower limit of the normal state voltage u n，min And the upper limit of the standard deviation of the node voltage in normal state σ n，max ; For the system historical operation data sample within any time window T, if the minimum node voltage u min >u n，min And the maximum standard deviation of the node voltage σ max <σ n,max, then the system historical operation data samples within the time window T are determined to be data samples in normal operating state; the data samples in normal operating state corresponding to N different time windows T constitute a data sample set.

[0026] Furthermore, u n，min and σ n，max It is related to factors such as the fuel cell stack structure and membrane electrode materials used.

[0027] Furthermore, the specific process of outlier processing in step 3 is as follows:

[0028] Step 3.1: Determine the sliding window width w based on the discharge period of impurity gas at the anode outlet o , let p = 1;

[0029] Step 3.2: For each operating parameter i of the pth training sample in the training data set p ,i=1,2,...,8, traverse to get p~p+w o Running parameter i in the sliding window j ,j=median i of p,p+1,...,p+w0 p,median ; Among them, 1, 2, ..., 8 represent current, cooling water inlet temperature, cooling water outlet temperature, anode inlet pressure, anode outlet pressure, hydrogen circulation flow, air flow and node voltage respectively;

[0030] Step 3.3: Calculate p~p+w o Each operating parameter i in the sliding window j Corresponding to i p,median The absolute value of the deviation Δi j :

[0031] Δi j =|i j -i p,median |,j=p,p+1,...,p+w0

[0032] Step 3.4: Calculate Δi j The median Δi p,median is the median absolute deviation;

[0033] Step 3.5: The data sample is determined to be an outlier, where n is the preset allowable deviation coefficient; the outlier is adjusted to the mean of the first non-outlier value on both sides of it. If the outlier is the endpoint of the training data set, it is adjusted to the closest non-outlier value;

[0034] Step 3.6: Determine whether p is equal to N. If not, set p = p + 1 and go to step 3.2; if so, end the outlier processing process.

[0035] Furthermore, n is 1-3.

[0036] Furthermore, the specific process of noise processing in step 4 is as follows: the moving window width w is determined based on the discharge period of the impurity gas at the anode outlet. s , the operating parameters i in the training data set after step 3 q ',i=1,2,...,8,q=1,2,...,N satisfies

[0037] Furthermore, the specific process of data standardization in step 5 is as follows: determine the lower limit a of the standardization interval and the upper limit b of the standardization interval, and the operating parameters i in the training data set processed in step 4 are q ',i=1,2,...,8,q=1,2,...,N satisfies

[0038] Furthermore, the lower limit of the normalized interval a=0, and the upper limit of the normalized interval b=1.

[0039] Furthermore, the desired output voltage in step 12 is By solving the FNN voltage model, the expected output node voltage sum is obtained, and the expected node voltage standard deviation It is obtained by solving the standard deviation of the expected output node voltage of the FNN voltage model.

[0040] The beneficial effects of the present invention are:

[0041] 1. This invention proposes a method for diagnosing water management faults in automotive fuel cell power systems. This method fully utilizes historical system operating data to train a data-driven model for water management fault diagnosis in PEMFC power systems. Without the need for additional equipment, this data-driven model replaces the mechanistic model, significantly reducing modeling difficulty and improving model accuracy and versatility. Based on empirical knowledge, a water management fault classification logic is constructed, and the residual between the data-driven model output and the actual operating output is used to rapidly diagnose normal operation, flooding, and drying faults.

[0042] 2. Based on the physical characteristics of the diagnostic indicators, the present invention selects FNN to build the node voltage model while taking into account both model complexity and model accuracy. Taking into account the strong time correlation of the anode pressure drop parameter, RNN is selected to build the anode pressure drop model. The regression performance and versatility of the above model are verified through experimental data and actual vehicle operation data. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a system block diagram of the existing water management fault diagnosis method based on the mechanism model;

[0044] Figure 2This is a system block diagram of the water management fault diagnosis method for a vehicle fuel cell power system proposed by the present invention;

[0045] Figure 3 This is a simplified structural diagram of the 30kW vehicle hydrogen fuel cell engine platform in Example 1 of the present invention;

[0046] Figure 4 Schematic diagram of the FNN voltage model in Example 1 of the present invention;

[0047] Figure 5 This is the iterative process of the FNN voltage model loss function in Example 1 of the present invention;

[0048] Figure 6 This is a graph showing the regression effect of the FNN voltage model in Example 1 of the present invention;

[0049] Figure 7 Schematic diagram of the RNN anode pressure drop model in Example 1 of the present invention;

[0050] Figure 8 This is the iterative process of the loss function of the RNN anode pressure drop model in Example 1 of the present invention;

[0051] Figure 9 This is a regression effect diagram of the RNN anode pressure drop model in Example 1 of the present invention;

[0052] Figure 10 This is a diagram of the water management fault diagnosis results in Example 1 of the present invention. DETAILED DESCRIPTION

[0053] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention is further described in conjunction with the following specific embodiments and with reference to the accompanying drawings.

[0054] The following non-limiting embodiments may enable those skilled in the art to more fully understand the present invention, but are not intended to limit the present invention in any way.

[0055] Example 1

[0056] This embodiment is based on Figure 3 The 30kW automotive hydrogen fuel cell engine platform is shown, and a water management fault diagnosis method for the automotive fuel cell power system is proposed.

[0057] The automotive hydrogen fuel cell engine platform can measure and store real-time system parameters, including anode inlet and outlet stack pressures, hydrogen circulation flow, cooling water inlet and outlet stack temperatures, air flow, air compressor speed, current, and node voltage. The data sampling frequency is 10 Hz. The automotive hydrogen fuel cell engine platform operates in hydrogen circulation and cathode exhaust direct discharge mode, with anode timed purge on and off cycles of 20 seconds and 0.5 seconds, respectively. The data sets involved in this example were acquired from the automotive hydrogen fuel cell engine platform.

[0058] The system block diagram of the vehicle fuel cell power system water management fault diagnosis method is as follows: Figure 2 As shown, the specific steps include:

[0059] Step 1: Filter the data sample set in normal operation state from the historical operation data of the vehicle hydrogen fuel cell engine platform system in the past month. The specific screening criteria are: set the time window T = 60s, the lower limit of the normal state voltage u n，min =0.65V, the upper limit of the standard deviation of the node voltage in normal state σ n，max =0.075mV; For the system historical operation data sample within any time window T, if the minimum node voltage u min >u n，min And the maximum standard deviation of the node voltage σ max <σ n,max , then the system historical operation data samples within the time window T are determined to be data samples in normal operating state; the data samples in normal operating state corresponding to N different time windows T constitute a data sample set;

[0060] Step 2: Select operating parameters from the data sample set to form a training data set; wherein the operating parameters include current, cooling water inlet temperature, cooling water outlet temperature, anode inlet pressure, anode outlet pressure, hydrogen circulation flow, air flow, and node voltage;

[0061] Step 3: Use the absolute median deviation method based on sliding window to process outliers in the training data set. The specific steps are as follows:

[0062] Step 3.1: Determine the sliding window width w based on the discharge period of impurity gas at the anode outlet o =205, let p=1;

[0063] Step 3.2: For each operating parameter i of the pth training sample in the training data set p ,i=1,2,...,8, traverse to get p~p+w o Running parameter i in the sliding window j ,j=median i of p,p+1,...,p+w0 p,median; Among them, 1, 2, ..., 8 represent current, cooling water inlet temperature, cooling water outlet temperature, anode inlet pressure, anode outlet pressure, hydrogen circulation flow, air flow and node voltage respectively;

[0064] Step 3.3: Calculate p~p+w o Each operating parameter i in the sliding window j Corresponding to i p,median The absolute value of the deviation Δi j :

[0065] Δi j =|i j -i p,median |,j=p,p+1,...,p+w0

[0066] Step 3.4: Calculate Δi j The median Δi p,median is the median absolute deviation;

[0067] Step 3.5: The data sample is determined as an outlier, where n = 3; the outlier is adjusted to the mean of the first non-outlier value on both sides of it. If the outlier is the endpoint of the training data set, it is adjusted to the closest non-outlier value;

[0068] Step 3.6: Determine whether p is equal to N. If not, set p = p + 1 and go to step 3.2. If yes, end the outlier processing process.

[0069] Step 4: Use the moving average smoothing filter method to perform noise processing on the training data set processed in step 3. The specific process is: determine the moving window width w based on the discharge period of the impurity gas at the anode outlet s =205, the operating parameters i in the training data set after step 3 q ',i=1,2,...,8,q=1,2,...,N satisfies

[0070] Step 5: Use the minimum-maximum standardization method to standardize the training data set processed in step 4. The specific process is: determine the lower limit of the standardization interval a = 0 and the upper limit of the standardization interval b = 1, and the operating parameters i in the training data set processed in step 4 are q ',i=1,2,...,8,q=1,2,...,N satisfies

[0071] Step 6: Select the first part of the operating parameters in the training data set processed in step 5 to form the voltage model input data set. The last input node is a random positive natural number c, which is recommended to be 1. The first part of the operating parameters includes current I, cooling water inlet temperature T coolant,in , Cooling water outlet temperature T coolant,out , anode inlet pressure P AN,in , hydrogen circulation flow Q loop and air flow Q air ;

[0072] Step 7: Select the node voltage in the training data set processed in step 5 as the voltage model output data set;

[0073] Step 8: Select the second part of parameters from the training data set processed in step 5 to form the anode pressure drop model input data set. The last input node is a random positive natural number d, which is recommended to be 1. The second part of the operating parameters includes current I, cooling water inlet temperature T coolant,in , Cooling water outlet temperature T coolant,out , anode inlet pressure P AN,in , hydrogen circulation flow Q loop ;

[0074] Step 9: Select the difference between the anode inlet pressure and the anode outlet pressure in the training data set processed in step 5 as the output data set of the anode pressure drop model;

[0075] Step 10: Take the voltage model input dataset described in step 6 as input and the voltage model output dataset described in step 7 as the expected output, and train the FNN based on the minimum gradient descent method to obtain the following: Figure 4 The FNN voltage model shown in Figure 2 is used as the training loss function, which is the mean of the Euclidean distance between the FNN output and the expected output. The iterative descent process of the loss function is as follows: Figure 5 As shown; the regression effect of FNN voltage model on the data set is as follows Figure 6 As shown;

[0076] Among them, the main parameters of the FNN voltage model are: the number of hidden neurons n F =8, loss function target value E F =1mV, the activation function is

[0077] Step 11: Using the anode pressure drop model input dataset described in step 8 as input and the anode pressure drop model output dataset described in step 9 as the expected output, train the RNN based on the minimum gradient descent method to obtain the following: Figure 7The RNN anode voltage drop model shown in Figure 2 is used as the training loss function. The training loss function is the mean of the Euclidean distance between the RNN output and the expected output. The iterative descent process of the loss function is as follows: Figure 8 As shown; the regression effect of the RNN anode voltage drop model on the data set is as follows Figure 9 As shown;

[0078] Among them, the main parameters of the RNN anode voltage drop model are: the number of hidden layer neurons n F =5, loss function target value E F =10Pa, the activation function is

[0079] Step 12: The FNN voltage model and the RNN anode voltage drop model are used in parallel to form a data-driven model. A water management fault classifier is constructed based on the residual between the output of the data-driven model and the actual output of the PEMFC system during operation. The output of the data-driven model includes the expected output voltage Expected node voltage standard deviation and the expected anode pressure drop Expected output voltage By solving the FNN voltage model, the expected output node voltage is obtained The expected node voltage standard deviation is obtained by summing By solving the FNN voltage model, the expected output node voltage is obtained The standard deviation of the actual output voltage U is obtained. The fuel cell used in this embodiment has 7 battery units. The actual output of the PEMFC system during operation includes the actual output voltage U, the actual unit voltage standard deviation U STD and actual anode pressure drop ΔP AN ;

[0080] Water management fault diagnosis is completed based on the water management fault classifier. The specific steps are as follows:

[0081] Step 12.1: Preset and Corresponding to U, U STD and ΔP AN The upper limit of normality of the residual i ;where, i={U,U STD ,ΔP AN};

[0082] Step 12.2: Calculate separately and Corresponding to U and U at the current moment STD and ΔP AN The residual in,

[0083] Step 12.3: Determine δU >l U and Is it true? If so, go to step 12.4; if false, output the diagnosis result as normal;

[0084] Step 12.4: Judgment Is it true? If it is true, the output diagnosis result is flooding; if it is false, the output diagnosis result is dry.

[0085] The diagnostic results of this embodiment in actual application are as follows Figure 10 shown.

[0086] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the purpose and scope of the technical solutions, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for diagnosing water management faults in a vehicle fuel cell power system, characterized in that: The following steps are involved: Step 1: Filter the data sample set in normal operation state from the system historical operation data; Step 2: Select operating parameters from the data sample set to form a training data set; wherein the operating parameters include current, cooling water inlet temperature, cooling water outlet temperature, anode inlet pressure, anode outlet pressure, hydrogen circulation flow, air flow, and node voltage; Step 3: Use the absolute median deviation method based on sliding window to deal with outliers in the training data set; Step 4: Use the moving average smoothing filter method to perform noise processing on the training data set processed in step 3; Step 5: Use the minimum-maximum normalization method to normalize the training data set processed in step 4; Step 6: Selecting the first part of the operating parameters in the training data set processed in step 5 to form a voltage model input data set; wherein the first part of the operating parameters includes current, cooling water inlet temperature, cooling water outlet temperature, anode inlet pressure, hydrogen circulation flow rate, and air flow rate; Step 7: Select the node voltage in the training data set processed in step 5 as the voltage model output data set; Step 8: Select the second part of parameters in the training data set processed in step 5 to form the anode pressure drop model input data set; wherein the second part of operating parameters includes current, cooling water inlet temperature, cooling water outlet temperature, anode inlet pressure, and hydrogen circulation flow rate; Step 9: Select the difference between the anode inlet pressure and the anode outlet pressure in the training data set processed in step 5 as the output data set of the anode pressure drop model; Step 10: Using the voltage model input dataset as input and the voltage model output dataset as the expected output, the FNN is trained based on the minimum gradient descent method to obtain the FNN voltage model; wherein the training loss function used is the mean Euclidean distance between the FNN output and the expected output; Step 11: Using the anode pressure drop model input dataset as input and the anode pressure drop model output dataset as the expected output, the RNN is trained based on the minimum gradient descent method to obtain the RNN anode pressure drop model; wherein the training loss function used is the mean Euclidean distance between the RNN output and the expected output; Step 12: The FNN voltage model and the RNN anode voltage drop model are used in parallel to form a data-driven model. Based on the residual between the output of the data-driven model and the actual output of the system operation process, a water management fault classifier is constructed; wherein the output of the data-driven model includes the expected output voltage Expected node voltage standard deviation and the expected anode pressure drop The actual output of the system operation process includes the actual output voltage U, the actual voltage standard deviation U STD and actual anode pressure drop ΔP AN ; Water management fault diagnosis is completed based on the water management fault classifier. The specific steps are as follows: Step 12.1: Preset and Corresponding to U, U STD and ΔP AN The upper limit of the normal value of the residual; where i={U,U STD ,ΔP AN }; Step 12.2: Calculate separately and Corresponding to U and U at the current moment STD and ΔP AN The residual in, Step 12.3: Determine ε U >l U and Is it true? If so, go to step 12.4; if false, output the diagnosis result as normal; Step 12.4: Judgment Is it true? If it is true, the output diagnosis result is flooding; if it is false, the output diagnosis result is dry.

2. The method for diagnosing water management faults in a fuel cell power system for a vehicle according to claim 1, characterized in that: The specific criteria for screening in step 1 are: setting the time window T, the lower limit of the voltage in the normal state u n，min And the upper limit of the standard deviation of the node voltage in normal state σ n，max ; For the system historical operation data sample within any time window T, if the minimum node voltage u min >u n，min And the maximum standard deviation of the node voltage σ max <σ n,max , then it is determined that the system historical operation data samples within the time window T are data samples in normal operating state; The data samples of normal operating status corresponding to N different time windows T constitute a data sample set.

3. The method for diagnosing water management faults in a fuel cell power system for a vehicle according to claim 2, characterized in that: The specific process of outlier processing in step 3 is as follows: Step 3.1: Determine the sliding window width w based on the discharge period of impurity gas at the anode outlet o , let p = 1; Step 3.2: For each operating parameter i of the pth training sample in the training data set p ,i=1,2,...,8, traverse to get p~p+w o Running parameter i in the sliding window j ,j=median i of p,p+1,...,p+w0 p,median ; Among them, 1, 2, ..., 8 represent current, cooling water inlet temperature, cooling water outlet temperature, anode inlet pressure, anode outlet pressure, hydrogen circulation flow, air flow and node voltage respectively; Step 3.3: Calculate p~p+w o Each operating parameter i in the sliding window j Corresponding to i p,median The absolute value of the deviation Δi j : Δi j =|i j -i p,median |,j=p,p+1,...,p+w0 Step 3.4: Calculate Δi j The median Δi p,median is the median absolute deviation; Step 3.5: The data sample is determined to be an outlier, where n is the preset allowable deviation coefficient; the outlier is adjusted to the mean of the first non-outlier value on both sides of it. If the outlier is the endpoint of the training data set, it is adjusted to the closest non-outlier value; Step 3.6: Determine whether p is equal to N. If not, set p = p + 1 and go to step 3.2; if so, end the outlier processing process.

4. The method for diagnosing water management faults in a fuel cell power system for a vehicle according to claim 3, characterized in that: 1 to 3.

5. The method for diagnosing water management faults in a vehicle fuel cell power system according to claim 2, characterized in that: The specific process of noise processing in step 4 is: determine the moving window width w based on the discharge period of impurity gas at the anode outlet s , the operating parameters i in the training data set after step 3 q ',i=1,2,...,8,q=1,2,...,N satisfies 6. The method for diagnosing water management faults in a fuel cell power system for a vehicle according to claim 2, characterized in that: The specific process of data standardization in step 5 is: determine the lower limit a of the standardization interval and the upper limit b of the standardization interval, and the operating parameters i in the training data set processed in step 4 are q ',i=1,2,...,8,q=1,2,...,N satisfies 7. The method for diagnosing water management faults in a fuel cell power system for a vehicle according to claim 6, characterized in that: The lower limit of the normalized interval a=0, and the upper limit of the normalized interval b=1.

8. The method for diagnosing water management faults in a fuel cell power system for a vehicle according to claim 1, characterized in that: The desired output voltage in step 12 By solving the FNN voltage model, the expected output node voltage sum is obtained, and the expected node voltage standard deviation It is obtained by solving the standard deviation of the expected output node voltage of the FNN voltage model.

Images