Fuel cell fault diagnosis method based on neural network

Abstract

A fuel cell fault diagnosis method based on a neural network. The method comprises: for a PEMFC system, acquiring operating state data, wherein the PEMFC system comprises a humidifier and a fuel cell stack; inputting the operating state data into a trained fault pre-diagnosis model, so as to predict a fault diagnosis result of the fuel cell stack at a subsequent moment. The fault pre-diagnosis model is constructed on the basis of a Transformer neural network, and comprises an encoder and a decoder, wherein the encoder comprises a plurality of small encoders, and each small encoder is provided with a self-attention layer and a feedforward neural network layer; and the decoder comprises a plurality of small decoders, and each small decoder comprises a self-attention layer, an attention layer and a feedforward neural network layer. An early warning can be provided for a water management fault of a PEMFC in an operating state, and the accuracy of pre-diagnosis is effectively improved.

Description

Fuel cell fault diagnosis method based on neural network Technical Field

[0001] The present invention relates to the technical field of fuel cell management, and more particularly to a fuel cell fault diagnosis method based on a neural network. Background Art

[0002] Proton Exchange Membrane Fuel Cell (PEMFC) converts the chemical energy contained in hydrogen into electrical energy. It has the advantages of being clean and pollution-free, having high power density and low operating noise, and is therefore widely used in fields such as automobiles.

[0003] The PEMFC is a complex, nonlinear system with multiple coupled physical fields and parameters. However, it currently suffers from low reliability and poor durability, with water management failure being one of the main causes. Water management failures can be categorized as membrane drying and flooding. Membrane drying occurs when the proton exchange membrane (PEM) loses its wetness, resulting in decreased proton conductivity and, consequently, poor PEMFC performance. Flooding, on the other hand, causes excessive moisture within the PEMFC stack, clogging the gas diffusion layer and hindering the proper transport of reactants, thereby reducing reaction efficiency and the voltage generated by the electrochemical reaction. Both types of failures can reduce PEMFC performance and operating life. PEMFC pre-diagnosis can provide early warning of water management failure trends and prompt measures to eliminate them, thereby ensuring trouble-free PEMFC operation. Therefore, selecting appropriate pre-diagnosis methods is crucial for the future industrialization of PEMFCs.

[0004] With the development of artificial intelligence and the advancement of computer technology, data-driven methods have received widespread attention in the field of PEMFC fault diagnosis. This method does not require the establishment of an accurate model. It directly uses the historical data of the PEMFC stack during operation as training data to obtain the corresponding intelligent algorithm model. The model determines the membrane dryness, flooding and normal status of the PEMFC stack based on the current operating data, which reduces the requirements for equipment compared to experimental test methods.

[0005] In the existing technology, the data-driven fault diagnosis method mainly includes four steps: data collection, feature extraction, data classification and diagnostic decision-making. The current innovation direction focuses on combining machine learning and other technologies to mine fault feature information. The machine learning methods currently widely used for fault diagnosis include neural networks, fuzzy clustering, neuro-fuzzy, random forests, Bayesian networks, support vector machines, etc. For example, patent application CN114048772A discloses a fault diagnosis method for a fuel cell device, which obtains a data set by collecting sensor signals under normal working conditions and sensor signals under fault conditions, inputs the training sample set into a constructed multi-layer long short-term memory network (Long Short-Term Memory, LSTM) for training, and obtains the optimal multi-layer LSTM network; the re-acquired test data set is input into the optimal multi-layer LSTM network to obtain the fault type of the test data set. Patent application CN111600051A discloses a PEMFC fault diagnosis method based on image processing. This method obtains one-dimensional membrane dry and flooding fault voltage data from the PEMFC system, converts the one-dimensional data into a two-dimensional image, extracts two image features from the two-dimensional image: the image grayscale decrease rate and the pixel grayscale upward fluctuation ratio. The image features are then classified and diagnosed using a K-means clustering algorithm. Patent application CN113962259A discloses a multi-mode, two-layer fault diagnosis method for PEMFC systems. This method uses typical fault operation data to train a support vector machine (SVM) fault diagnosis algorithm, classifying fault types into low-pressure faults, high-pressure faults, membrane dry faults, and low air excess coefficient faults. Finally, a cross-validation method is used to verify the accuracy of the model detection.

[0006] In summary, the current application of PEMFC still faces the problems of low reliability and poor durability. Exploring accurate and effective fault diagnosis methods is of great significance to the future development of industrialization. Existing PEMFC fault diagnosis usually uses machine learning methods to achieve fault identification. Although it has the advantages of fast diagnosis speed and high diagnostic accuracy, the diagnosis process is mainly applied after the PEMFC fault occurs and cannot provide early warning of the fault status. In addition, since the existing solutions cannot monitor the water content inside the PEMFC during use, and the irreversible damage caused by water management failure will affect the operating life of the PEMFC, pre-diagnosis is of great significance to reducing the life degradation caused by PEMFC faults.

[0007] Summary of the Invention

[0008] The purpose of the present invention is to overcome the above-mentioned shortcomings of the prior art and provide a fuel cell fault diagnosis method based on a neural network. The method comprises:

[0009] Collecting operating status data for a PEMFC system, wherein the PEMFC system includes a humidifier and a fuel cell stack;

[0010] Inputting the operating status data into a trained fault pre-diagnosis model to predict a fault diagnosis result of the fuel cell stack at a subsequent time;

[0011] Among them, the fault prediction model is constructed based on the Transformer neural network, including an encoder and a decoder. The encoder contains multiple small encoders, each of which is provided with a self-attention layer and a feedforward neural network layer; the decoder contains multiple small decoders, each of which contains a self-attention layer, an attention layer and a feedforward neural network layer.

[0012] Compared with the existing technology, the advantage of the present invention is that it uses a Transformer neural network to construct a fault pre-diagnosis model. This model is suitable for the prediction of nonlinear systems and aims to provide early warning of PEMFC water management failures in operation. It also effectively improves the accuracy of pre-diagnosis, and then eliminates the failure through subsequent measures, thereby achieving the goal of trouble-free operation of PEMFC.

[0013] Further features and advantages of the present invention will become apparent from the following detailed description of exemplary embodiments of the present invention with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and, together with the description, serve to explain the principles of the invention.

[0015] FIG1 is a flow chart of a fuel cell fault diagnosis method based on a neural network according to an embodiment of the present invention;

[0016] FIG2 is a schematic diagram of a proton exchange membrane fuel cell stack test according to one embodiment of the present invention;

[0017] FIG3 is a flow chart of water management fault characterization according to one embodiment of the present invention;

[0018] FIG4 is a schematic diagram of a framework of a proton exchange membrane fuel cell fault prediction model constructed based on a Transformer neural network according to one embodiment of the present invention;

[0019] FIG5 is a flow chart of a proton exchange membrane fuel cell fault diagnosis strategy based on a Transformer neural network according to an embodiment of the present invention. DETAILED DESCRIPTION

[0020] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that unless otherwise specifically stated, the relative arrangement of components and steps, numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present invention.

[0021] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the invention, its application, or uses.

[0022] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered part of the specification.

[0023] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not limiting. Therefore, other examples of the exemplary embodiments may have different values.

[0024] It should be noted that like reference numerals and letters refer to like items in the following figures, and therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0025] The present invention optimizes the fault diagnosis method of the PEMFC system, advances the diagnosis to a period of time before the fault occurs, can provide early warning of the fault trend, and then take timely measures to eliminate it, thereby achieving trouble-free operation of the PEMFC system.

[0026] In general, the present invention involves performing two sets of PEMFC stack tests and characterizing water management failures. These two sets of PEMFC stack tests are conducted under identical conditions, obtaining data including PEMFC stack temperature, anode and cathode humidity, voltage, and cathode pressure drop. The PEMFC voltage and cathode inlet and outlet pressure drop obtained from the first set of tests characterize water management failures. A PEMFC fault pre-diagnosis model is then constructed based on a Transformer neural network. The second set of test data is divided into a training set and a test set. The training set is used to train the fault pre-diagnosis model and obtain optimized model parameters, while the test set is used to evaluate the training results of the fault pre-diagnosis model.

[0027] 1 , the provided fuel cell fault diagnosis method based on a neural network includes the following steps:

[0028] Step S110 , collecting a sample data set, and analyzing the corresponding relationship between the operating state characteristic data of the PEMFC system and the fuel cell fault, and then constructing a training set.

[0029] To collect a sample data set, a PEMFC stack test bench was used for testing, as shown in Figure 2. This test bench utilizes a humidifier, air handling system, and hydrogen handling system to control the operating conditions of the PEMFC stack (or fuel cell, for short). A pressure sensor was used to monitor the operating status. For example, during testing, the PEMFC stack operating conditions were controlled to observe the relationship between stack membrane dryness and flooding failures and voltage and cathode pressure drop. Membrane dryness frequently occurs under low humidity conditions, while flooding is even more frequent under high humidity conditions. By setting initial test conditions such as temperature and humidity, the voltage and cathode pressure drop were tested in multiple steps and observed to determine their relationship to membrane dryness and flooding. See Table 1 for sample data examples. The sample data can be arranged in chronological order.

[0030] Table 1 PEMFC stack sample data

[0031] When a PEMFC is in operation, a dry membrane failure occurs when the PEM (proton exchange membrane) loses its moisture content. This reduces proton conductivity and prevents effective proton transfer across the membrane, resulting in a drop in voltage. However, a dry membrane failure does not significantly alter the cathode pressure drop, as this is primarily influenced by damping and gas diffusion through the conductive material, and is largely unrelated to the presence of water. A flooding failure results in excessive moisture within the PEMFC stack, clogging the gas diffusion layer and hindering the proper transport of reactants, thereby reducing reaction efficiency and the voltage generated by the electrochemical reaction. Furthermore, the accumulated water increases cathode humidity, leading to an increase in cathode pressure drop.

[0032] Based on the voltage corresponding to the water management fault and the cathode pressure drop variation pattern described above, the water management fault type corresponding to the labeled data in Table 1 is determined. For example, the output voltage of the PEMFC stack is compared with the ideal output voltage to determine whether a fault has occurred. If the output voltage exceeds the normal range, it indicates a PEMFC stack fault. The cathode inlet and outlet pressure drops are then compared with the normal pressure drop to determine the fault type under a given power demand. If the pressure drop exceeds the normal range, it can be determined as a flooding fault; if it is within the normal range, it can be determined as a membrane dry-out fault. The water management fault characterization process is shown in Figure 3.

[0033] By testing the PEMFC stack, a sample data set can be obtained. Using the voltage and cathode pressure drop data from the sample data set to characterize the relationship between PEMFC stack failures, a training set can be constructed. This training set reflects the correspondence between operating status data and failure types. It should be understood that the failure types described herein include both specific fault signal types and normal states.

[0034] Step S120: constructing a fault pre-diagnosis model based on the Transformer neural network.

[0035] Transformer is a model that uses the attention mechanism to improve training speed. The training process of its basic units is parallel computing, which is more suitable for training in large-scale distributed clusters. When performing time series prediction, this model can capture the global dependency pattern of the original time series signal through the self-attention mechanism, significantly improving diagnostic accuracy and generalization performance.

[0036] In one embodiment, a PEMFC fault pre-diagnosis model constructed based on Transformer is shown in FIG4 . The fault pre-diagnosis model mainly includes an encoder and a decoder module, which adopts a sequence-to-sequence (seq2seq) model structure. The model input is divided into k non-overlapping time series according to the time dimension and a one-dimensional vector is constructed, which is linearly mapped to the model output. The encoder is composed of N identical small encoders, each of which contains a self-attention layer and a feedforward neural network layer. After the data is processed by the self-attention layer, it is input into a fully connected feedforward neural network. The decoder is also composed of N identical small decoders, each of which contains two self-attention layers and a feedforward neural network layer, and the first self-attention sublayer is masked to ensure that the prediction of position i can only depend on the output of positions less than i. The feedforward neural network can, for example, adopt a fully connected network type.

[0037] The self-attention module uses the Query-Key-Value (QKV) model, and its function can be described as mapping a Q vector and a set of (K, V) vector pairs to an output vector. The input consists of a Q vector and a K vector, and the output is calculated as a weighted sum of V vectors. The weight assigned to each V vector is derived from a similarity function between the Q vector and the corresponding K vector. In practice, the attention function is calculated simultaneously on a set of Q vectors, which are packaged into a matrix Q. The K vectors and V vectors are also packaged into matrices K and V. The calculated output is as follows:

[0038] Among them, softmax is the normalization function, is a scaling factor that prevents the normalization function’s input value from being too large, causing the partial derivative to approach zero.

[0039] Figure 4 shows a fault pre-diagnosis model that obtains fault pre-diagnosis results based on input data. First, the labeled data voltage and cathode pressure are converted into water management fault signals (or fault types). Then, the feature data obtained during testing and the fault signals are combined into a training set. The feature data is input in a time-series manner, and the fault signals are used as the model output. The fault pre-diagnosis model is trained in a single step. For example, the input data includes the PEMFC stack operating status and other relevant data, such as PEMFC stack temperature, anode and cathode humidity, voltage, and cathode pressure drop, to capture the relationship between water management faults and various characteristic parameters. The encoder is responsible for mapping the input sequence into a set of high-dimensional vector representations. The encoder utilizes a self-attention mechanism, enabling the model to simultaneously focus on different positions in the input sequence and capture corresponding relationships within the sequence. The encoder output is passed to the decoder, which receives the encoder output and generates a PEMFC pre-diagnosis result based on it. The decoder dynamically weights the input sequence through a self-attention mechanism and attention to the encoder output, generating a representation relevant to fault pre-diagnosis. The fault pre-diagnosis result is the water management fault pre-diagnosis result after a period of time. For example, the pre-diagnosis result includes membrane dry fault, water flooding fault and normal state.

[0040] Step S130: training a fault pre-diagnosis model using the training set.

[0041] In one embodiment, a training set and a test set may be constructed based on the sample data set to train and evaluate the fault pre-diagnosis model.

[0042] The fault prediction model is a combination of an encoder and a decoder, which maps the operating data of the PEMFC stack as an input sequence to the fault prediction results, where the encoder captures the feature representation of the input sequence and the decoder generates the diagnosis results based on these representations.

[0043] As shown in FIG6 , the process of training the PEMFC fault prediction model using the training set includes the following steps:

[0044] S1: Sort the PEMFC stack operating data in time series and import it into the encoder, and add position coding. The PEMFC stack operating data includes humidity, temperature, current density and other data.

[0045] S2: The self-attention mechanism and fully connected feedforward neural network in the encoder process the data, and the encoder output value is obtained through the self-attention mechanism.

[0046] S3: Repeat step S2 N times (N is the number of small encoders), and import the obtained data into the decoder.

[0047] S4: Train and predict the results in the decoder until the set termination condition is met, otherwise continue training and prediction.

[0048] After the training is completed, the performance of the fault prediction model can be evaluated using the set evaluation index. For example, the accuracy A=I ac / I total As an evaluation indicator, I ac Indicates the number of faults diagnosed correctly, I total Represents the total number of faults in the training set. Furthermore, the test set can be used to further test the performance of the fault prediction model. If the performance does not meet the requirements, a new training set can be constructed to fine-tune or retrain the model.

[0049] Step S140 : For the PEMFC system in operation, a fault early warning is performed using the trained fault pre-diagnosis model.

[0050] After the fault pre-diagnosis model is trained, real-time fault prediction can be performed. For example, the PEMFC system's operating status characteristic data is collected and input into the trained fault pre-diagnosis model to obtain fault diagnosis results at subsequent moments. If a fault is predicted to occur, a fault warning is issued so that timely troubleshooting measures can be taken.

[0051] It should be noted that the model training process described in the present invention can be performed offline on a server or in the cloud, and real-time fault diagnosis or prediction can be achieved by embedding the trained model into an electronic device. The electronic device can be a terminal device or a server, and the terminal device includes any terminal device such as a mobile phone, tablet computer, personal digital assistant (PDA), point of sale (POS), vehicle-mounted computer, smart wearable device, etc. The server includes but is not limited to an application server or a web server, and can be a standalone server, a cluster server, or a cloud server.

[0052] In summary, the present invention optimizes the PEMFC fault diagnosis process and has at least the following advantages over the prior art:

[0053] 1) The present invention utilizes a variety of collectible parameter signals and develops a method for pre-diagnosis of PEMFC water management failures based on a deep learning Transformer algorithm, which can provide early warning of failure trends.

[0054] 2) The present invention uses a Transformer neural network to construct a fault pre-diagnosis model, which is suitable for the prediction of nonlinear systems. It can pre-diagnose water management faults in PEMFC and effectively improve the accuracy of pre-diagnosis. Then, through subsequent measures to eliminate the faults, the goal of trouble-free operation of PEMFC can be guaranteed.

[0055] 3) It has been verified that the present invention can not only provide accurate fault warning for fuel cells, but can also be extended to fault pre-diagnosis of similar systems.

[0056] The present invention may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present invention.

[0057] Computer-readable storage medium can be a tangible device that can keep and store the instructions used by the instruction execution device.Computer-readable storage medium can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device or any suitable combination thereof.More specific examples (non-exhaustive list) of computer-readable storage medium include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, for example, a punch card or a convex structure in a groove having instructions stored thereon, and any suitable combination thereof.Computer-readable storage medium used herein is not interpreted as a transient signal itself, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagated by waveguides or other transmission media (for example, light pulses by fiber optic cables), or electrical signals transmitted by wires.

[0058] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions to be stored in the computer-readable storage medium in each computing / processing device.

[0059] The computer program instructions for performing the operation of the present invention can be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, Python, and conventional procedural programming languages ​​such as "C" language or similar programming languages. The computer readable program instructions can be executed entirely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer, partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), is personalized by utilizing the state information of the computer readable program instructions, and the electronic circuit can execute the computer readable program instructions, thereby realizing various aspects of the present invention.

[0060] Various aspects of the present invention are described herein with reference to flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present invention. It should be understood that each block of the flowcharts and / or block diagrams, and combinations of blocks in the flowcharts and / or block diagrams, can be implemented by computer-readable program instructions.

[0061] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine, so that when these instructions are executed by the processor of the computer or other programmable data processing device, a device is generated that implements the functions / actions specified in one or more blocks in the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium, where these instructions cause the computer, programmable data processing device, and / or other device to operate in a specific manner. Thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing various aspects of the functions / actions specified in one or more blocks in the flowchart and / or block diagram.

[0062] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operational steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to implement the functions / actions specified in one or more blocks in the flowchart and / or block diagram.

[0063] The flowcharts and block diagrams in the accompanying drawings show the possible implementation architecture, functions and operations of the systems, methods and computer program products according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, program segment or part of an instruction, and the module, program segment or part of the instruction contains one or more executable instructions for implementing the specified logical function. In some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system that performs the specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions. It is well known to those skilled in the art that implementation by hardware, implementation by software, and implementation by a combination of software and hardware are all equivalent.

[0064] While various embodiments of the present invention have been described above, the foregoing description is intended to be illustrative, non-exhaustive, and not limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is selected to best explain the principles of the embodiments, their practical applications, or technological improvements in the marketplace, or to enable others skilled in the art to understand the embodiments disclosed herein. The scope of the present invention is defined by the appended claims.

Claims

1. A fuel cell fault diagnosis method based on a neural network, comprising the following steps: For a PEMFC system, collecting characteristic data of an operating state, wherein the PEMFC system includes a humidifier and a fuel cell stack; Inputting the characteristic data of the operating state into a trained fault pre-diagnosis model to predict the fault diagnosis result of the fuel cell stack at a subsequent time; Among them, the fault prediction model is constructed based on the Transformer neural network, including an encoder and a decoder, the encoder includes multiple small encoders, each small encoder is provided with a self-attention layer and a feedforward neural network layer; the decoder includes multiple small decoders, each small decoder includes a self-attention layer, an attention layer for the output of the encoder, and a feedforward neural network layer.

2. The method according to claim 1, characterized in that The fault prediction model is trained according to the following steps: Collecting a sample data set, each sample reflecting a correspondence between operating state characteristic data and label data at multiple historical moments, wherein the label data includes a fuel cell stack voltage and an inlet and outlet pressure drop of a cathode of the fuel cell stack; Determine a corresponding fault type based on the label data, and construct a training set, which reflects the corresponding relationship between the operating characteristic data and the fault type; Inputting the running status feature data of the training set into the encoder in time series order and adding position coding; The self-attention layer and feedforward neural network of the encoder process the data and obtain the output data of the encoder through the self-attention mechanism; The output data of the encoder is imported into the decoder for training and predicting the fault diagnosis result until the set termination condition is reached.

3. The method according to claim 1, characterized in that: The operating status characteristic data includes one or more of the current information of the fuel cell stack, the current density information of the fuel cell stack, the temperature information of the fuel cell stack, the pressure information at the inlet and outlet of the fuel cell stack, the gas flow rate information at the inlet and outlet of the fuel cell stack, the cooling water flow rate information of the fuel cell stack and the air relative humidity information of the fuel cell stack.

4. The method according to claim 1, characterized in that The fault diagnosis results include membrane dry fault, water flooding fault and normal state.

5. The method according to claim 1, characterized in that Also includes: Using a test data set to test whether the performance of the trained fault pre-diagnosis model meets a predetermined evaluation index; If not, construct a new training set to fine-tune the fault prediction model; The evaluation index is set as: A=I ac / I total Among them, A represents the accuracy, I ac Indicates the number of faults diagnosed correctly, I total Represents the total number of faults in the test dataset.

6. The method according to claim 1, characterized in that The feedforward neural network is a fully connected feedforward neural network.

7. The method according to claim 2, characterized in that The self-attention mechanism is expressed as: Among them, Q represents the query matrix, K represents the key matrix, V represents the value matrix, and softmax is the normalization function. is the scaling factor.

8. The method according to claim 2, characterized in that: Follow these steps to determine the type of fault: Compare the output voltage of the fuel cell stack with the ideal output voltage. If the output voltage is not within the set normal range, it is determined that there is a fault. When it is determined that a fault exists, under a given power demand, the inlet and outlet pressure drops of the cathode of the fuel cell stack are compared with the set normal pressure drop range. If the pressure drop exceeds the set normal pressure drop range, it is determined to be a flooding fault. If the pressure drop is within the set normal pressure drop range, it is determined to be a membrane dry fault.

9. A computer-readable storage medium having a computer program stored thereon, wherein: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

10. A computer device comprising a memory and a processor, wherein a computer program capable of running on the processor is stored in the memory, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

Images