Fuel cell operation fault diagnosis method and device, medium and product

Abstract

The invention discloses a fuel cell operation fault diagnosis method and device, a medium and a product, and relates to the field of battery diagnos.The method comprises the steps that target working condition data of a target fuel cell are obtained in real time, and the target working condition data comprise a current load current value and current impedance spectrum data; taking the target working condition data as the input of a target deep learning model, and predicting the fault state of the target fuel cell through the target deep learning model; wherein the target deep learning model obtains the mapping relation between the target working condition data and the fault state through training learning, and the current load current value is introduced as an additional prediction basis, so that the prediction accuracy of the fault working condition can be improved; the problem that an existing fuel cell operation fault diagnosis method is insufficient in diagnosis precision is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of battery diagnosis, in particular to a fuel cell operation fault diagnosis method, device, medium and product. BACKGROUND

[0002] Due to the internal gas-water-heat-electricity multi-field coupling of the fuel cell, the internal state evolution is complex and changeable, and at the same time, due to the closed structure, it is difficult for the traditional monitoring method to directly obtain the internal state. Therefore, the lack of accurate internal operation state feedback brings difficulties to the real-time health management and control of the fuel cell.

[0003] In view of the internal state online monitoring problem, electrochemical impedance spectroscopy is widely used for dynamic characteristic analysis of electrochemical systems. By injecting different frequency disturbances, different frequency impedance responses are obtained for analyzing the current operation state of the system, which can reflect the information that cannot be directly measured by external physical monitoring. With the development of deep learning models, the mapping relationship between the electrochemical impedance spectroscopy of the fuel cell and the fault can be learned through the deep learning model, so that the deep learning model can predict the fault based on the electrochemical impedance spectroscopy of the fuel cell. However, the fault of the fuel cell is not only related to the electrochemical impedance spectroscopy, so the above fuel cell operation fault diagnosis method has the problem of insufficient diagnosis accuracy. SUMMARY

[0004] The purpose of the present application is to provide a fuel cell operation fault diagnosis method, device, medium and product, which can improve the prediction accuracy of the fault working condition of the fuel cell, and solve the problem of insufficient diagnosis accuracy of the current fuel cell operation fault diagnosis method.

[0005] To achieve the above purpose, the present application provides the following solutions: In a first aspect, the present application provides a fuel cell operation fault diagnosis method, comprising: obtaining target working condition data of a target fuel cell in real time, the target working condition data comprising a current load current value and current impedance spectroscopy data; inputting the target working condition data into a target deep learning model, and predicting a fault state of the target fuel cell through the target deep learning model, wherein the target deep learning model has learned a mapping relationship between the target working condition data and the fault state through training.

[0006] In a second aspect, the present application provides a computer device, comprising a memory, a processor, a computer program stored on the memory and executable on the processor, and the processor executes the computer program to implement the steps of the fuel cell operation fault diagnosis method according to any one of the above.

[0007] In a third aspect, the present application provides a computer readable storage medium, having stored thereon a computer program, which, when executed by a processor, implements the steps of the fuel cell operation fault diagnosis method according to any one of the preceding aspects.

[0008] In a fourth aspect, the present application provides a computer program product comprising a computer program, which, when executed by a processor, implements the steps of the fuel cell operation fault diagnosis method according to any one of the preceding aspects.

[0009] According to the embodiments provided in the present application, the following technical effects are disclosed. The present application provides a fuel cell operation fault diagnosis method, device, medium and product, which not only takes the current impedance spectrum data of the target fuel cell as a prediction basis, but also takes the current load current value of the target as a prediction basis, and simultaneously takes the current impedance spectrum data and the current load current value of the target fuel cell as the input of the target deep learning model, so that the target deep learning model can predict the fault state of the target fuel cell based on the current impedance spectrum data and the current load current value at the same time, and the prediction accuracy of the fuel cell fault working condition can be improved, and the problem of insufficient diagnosis precision of the current fuel cell operation fault diagnosis method is solved. BRIEF DESCRIPTION OF DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor based on these drawings.

[0011] Figure 1 It is an application environment diagram of a fuel cell operation fault diagnosis method in an embodiment of the present application; Figure 2 It is a flowchart of a fuel cell operation fault diagnosis method in an embodiment of the present application; Figure 3 It is an impedance spectrum diagram of a sample fuel cell corresponding to different fault states at different current densities in an embodiment of the present application; Figure 4 It is an architecture diagram of a deep learning network in an embodiment of the present application; Figure 5 It is an architecture diagram of a long short-term memory network in an embodiment of the present application; Figure 6 It is an application diagram of a fuel cell operation fault diagnosis method in an embodiment of the present application; Figure 7 It is a structure diagram of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0012] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative effort belong to the scope of protection of the present application.

[0013] The above purposes, features and advantages of the present application will be more obvious and easy to understand. The present application will be described in further detail below with reference to the drawings and specific embodiments.

[0014] The fuel cell operation fault diagnosis method provided in the embodiments of the present application can be applied to the application environment as shown in Figure 1 . The terminal 101 communicates with the server 102 through a network. The data storage system can store data required to be processed by the server 102. The data storage system can be separately arranged, or integrated on the server 102, or placed on a cloud or other server. The terminal 101 can send the target working condition data of the target fuel cell to the server 102. The server 102 takes the target working condition data as the input of the target deep learning model, and predicts the fault state of the target fuel cell through the target deep learning model. The server 102 can feed back the obtained fault state of the target fuel cell to the terminal 101. In addition, in some embodiments, the fuel cell operation fault diagnosis method can also be implemented by the server 102 or the terminal 101 alone, for example, the terminal 101 can directly take the real-time acquired target working condition data as the input of the target deep learning model, and predict the fault state of the target fuel cell through the target deep learning model. It should be noted that, in order to improve the calculation real-time performance, the fuel cell operation fault diagnosis method is preferentially executed by the terminal 101 deployed locally.

[0015] The terminal 101 can be, but is not limited to, various desktop computers, notebook computers, smart phones, tablet computers and smart vehicle-mounted devices. The server 102 can be implemented by an independent server or a server cluster composed of multiple servers, and can also be a cloud server.

[0016] In an exemplary embodiment, as shown in Figure 2 , a fuel cell operation fault diagnosis method is provided. The method is executed by a computer device, specifically, can be executed by a terminal or a server, or can be executed by a terminal and a server. In the embodiments of the present application, the method is applied to the terminal 101 (such as a smart vehicle-mounted device) in Figure 1 . The method includes the following steps 210 and 220.

[0017] In step 210, target working condition data of the target fuel cell is acquired in real time, and the target working condition data includes a current load current value (time domain data) and current impedance spectrum data (frequency domain data).

[0018] In step 220, the target working condition data is taken as an input of a target deep learning model, and a fault state of the target fuel cell is predicted by the target deep learning model; wherein the target deep learning model has learned a mapping relationship between the target working condition data and the fault state through training.

[0019] The target fuel cell is a fuel cell that needs to be detected for faults. In the above steps, not only the current impedance spectrum data of the target fuel cell is taken as a prediction basis, but also the current load current value of the target fuel cell is taken as a prediction basis. At the same time, the current impedance spectrum data and the current load current value of the target fuel cell are taken as inputs of the target deep learning model, so that the target deep learning model can predict the fault state of the target fuel cell based on the current impedance spectrum data and the current load current value at the same time. In a conventional prediction method, a deep learning model usually predicts faults of a fuel cell based on impedance spectrum data of the fuel cell. In view of the fact that the fuel cell has different impedance spectrum data in different load states under the same fault working condition, the fuel cell may have the same or similar impedance spectrum data when the fault working condition is A and the load state is a, and when the fault working condition is B and the load state is b. Therefore, it is difficult to accurately predict the fault working condition of the fuel cell only by relying on the impedance spectrum data. However, by introducing the current load current value as an additional prediction basis, the prediction accuracy of the fault working condition of the fuel cell can be improved, and the problem of insufficient diagnostic precision in the current fuel cell operation fault diagnosis method is solved.

[0020] Specifically, the current impedance spectrum data includes real part values and imaginary part values of impedance points in the current impedance spectrum; and the target working condition data is a data sequence, and elements of the data sequence include the current load current value, the real part values of the impedance points in the current impedance spectrum, and the imaginary part values of the impedance points in the current impedance spectrum.

[0021] In this embodiment, the target working condition data is input into the target deep learning model in a sequence form, so that the model can better process the target working condition data. For example, the current load current value can be taken as a first element in the data sequence, the real part values of the impedance points in the current impedance spectrum can be taken as a group of elements arranged after the current load current value in the data sequence, and the imaginary part values of the impedance points in the current impedance spectrum can be taken as a group of elements arranged after the real part values in the data sequence.

[0022] Specifically, the fault state includes the occurrence degree of various fault conditions, and different fault conditions include membrane dryness, flooding and gas starvation. Membrane dryness, flooding and gas starvation are three common fault conditions of fuel cells. In this embodiment, not only the type of fault condition occurring in the fuel cell is predicted, but also the occurrence degree of each fault condition is predicted. For example, in order to better quantify the output fault condition type and its occurrence degree, the occurrence degree of each fault condition is set to range from 0 to 1. 0 represents no fault condition of this type, and 1 represents that the fault condition is extremely serious. The closer the output is to 0, the lower the degree of the fault condition. On the contrary, the closer the output is to 1, the more serious the fault condition.

[0023] As shown in Figure 3 . The occurrence degree of the three fault conditions is simultaneously output as the model output, which means that some fault conditions can occur at the same time, which is consistent with the actual situation of the fuel cell, for example, flooding and gas starvation, gas starvation and membrane dryness of the fuel cell can occur at the same time. Relative to the labeled fault condition output, simultaneously outputting the occurrence degree of various fault conditions is helpful for subsequent control decisions.

[0024] Specifically, the training steps of the target deep learning model include: simulating different fault states of the sample fuel cell by adjusting the operating conditions of the sample fuel cell; obtaining impedance spectrum data of the sample fuel cell with different load current values under different fault states respectively; training the target deep learning model by taking the load current value and the impedance spectrum data as sample features and taking the fault state as sample labels.

[0025] In this embodiment, different values of operating conditions are set to simulate various fault states of the sample fuel cell, and then each fault state is traversed to obtain impedance spectrum data of the sample fuel cell with different load current values under each fault state through experiments. The impedance spectrum data can be obtained based on the stack test bench and impedance test equipment, and the operating conditions include air excess coefficient, water outlet temperature, anode and cathode pressure, and relative air humidity.

[0026] For example, based on the control variable method, four representative current density points (current density multiplied by the cross section of the load circuit is the load current) are taken from low to high current density (0.4 A / cm 2 ) to high current density (1.6 A / cm 2 ) at equal intervals. This current density covers the main working load range in the actual operation of the fuel cell system. At these four current density points (current density points), impedance spectra under different air excess coefficients, water outlet temperatures, anode and cathode pressures, and relative air humidities are measured. Table 1 shows the standard operating conditions under different current density points in this experiment.

[0027] Table 1 Standard operating conditions of fuel cells

[0028] At each current density point, the change control experiments of air excess coefficient, water outlet temperature, anode and cathode pressure and relative air humidity with the basic operating conditions as the intermediate values were carried out in turn. Table 2 is four groups of control experiments carried out at different current density points, each experiment taking air excess coefficient, water outlet temperature, anode and cathode pressure and relative air humidity as the independent variable, and one specific value combination of the independent variable corresponds to one fault state, and the non-independent variable parameters in each experiment are set according to the standard operating conditions.

[0029] Table 2 Parameter settings of control experiments of operating condition sensitivity of fuel cell

[0030] Before each impedance spectrum test, the operating conditions of the sample fuel cell were set by the host computer according to table 2, and then impedance spectrum measurement was carried out after the fuel cell worked stably.

[0031] The impedance test frequency covers the AC disturbance signal injection of 0.1Hz to 1000Hz, and the impedance points are selected for testing according to the octave, so as to obtain the impedance spectrum data of each current density point under various fault conditions between 0.1Hz and 1000Hz.

[0032] Specifically, the target deep learning model is a hybrid neural network model (which can be called a Transformer-LSTM fault diagnosis model), the hybrid neural network model includes a position encoding layer, a Transformer encoder, an LSTM decoder and a prediction layer, the Transformer encoder includes a multi-head self-attention network layer; the position encoding layer is used for position encoding of the target working condition data; the multi-head self-attention network layer is used for feature extraction and fusion of the position encoded target working condition data to generate the output of the Transformer encoder; the LSTM decoder is used for decoding the output of the Transformer encoder; and the prediction layer is used for predicting the fault state according to the output of the LSTM decoder.

[0033] The fuel cell operation fault diagnosis method in the embodiment is essentially a regression prediction problem. For such a problem, the embodiment adopts a hybrid neural network model with a Transformer plus LSTM architecture as the target deep learning model. By introducing a multi-head self-attention mechanism, the limitations of the model in capturing long-range dependencies in long sequences are improved, enabling it to better handle both local and global dependencies in the target working condition data sequence. In the embodiment, by combining the sequence prediction advantages of the LSTM decoder and the built-in transformation algorithm of the Transformer encoder, the connection between each impedance point in the regression model is enhanced, enabling the LSTM decoder to better handle complex sequence data tasks and improve computational efficiency.

[0034] With reference to Figure 4 , the target working condition data in the form of a sequence is used as the input sequence of the hybrid neural network model. The input sequence includes the load current value and the real and imaginary parts of each impedance point in the impedance spectrum data. The load current value is obtained in real time from the system current, and the real and imaginary parts of each impedance point are determined based on the real-time impedance spectrum data. After the input sequence is normalized by the Softmax function, a position encoding layer is used to generate a learnable embedding vector for each position of the input sequence, which displays the position encoding information. Then, the input sequence and the position encoding information are added to enhance the model's ability to perceive the positional relationship of each impedance point.

[0035] Exemplarily, for the input sequence , the calculation process of the Softmax function normalization is as follows:

[0036] wherein, n denotes the total length of the input sequence, and denote the i-th and j-th elements in the input sequence, respectively.

[0037] Further, a multi-head self-attention network layer is used to capture the global dependency between the input sequences. The self-attention mechanism is the core mechanism of the Transformer encoder, which dynamically aggregates global information by calculating the interaction weights (attention scores) between each element and other elements in the input sequence. The calculation principle of the self-attention layer is as follows: For single-head self-attention mechanism: First, the input sequence is linearly transformed as follows:

[0038] wherein, is the query matrix, is the key matrix, is the value matrix, is the position encoded input sequence, n is the total length of the input sequence, d is the feature dimension of the input at each position. , and are the learnable parameter weight matrices during the training process, is the feature dimension of the matrix and is the feature dimension of the matrix .

[0039] Secondly, the self-attention score is calculated:

[0040] where, is the attention score matrix, represents the similarity after the dot product of the query matrix and the key matrix, is the scaling factor to prevent the dot product value from being too large to cause gradient disappearance.

[0041] Then, the attention weight matrix is calculated:

[0042] In the above formula, the self-attention score is normalized to a probability distribution using the operation, which represents the degree of “attention” of each position to other positions.

[0043] Finally, the weighted sum of the attention weight matrix elements and the value matrix elements is obtained to obtain the output matrix:

[0044] where, is the output of the single-head self-attention mechanism, which represents the feature representation aggregated by global dependency relationships at each position.

[0045] For the multi-head self-attention mechanism, the linear transformation is modified as:

[0046] where, , , are the query matrix, key matrix, and value matrix of the i-th head, respectively, h is the number of heads. Then, the output matrix ​The output matrix of each head is represented as

[0047] wherein, is an output weight matrix, is an output matrix of the h th head, is an output of the multi-head self-attention mechanism, i.e., an output of the multi-head self-attention network layer.

[0048] It should be noted that in the present embodiment, two layers of multi-head self-attention network layers are preferably adopted. Referring to Figure 4 , the position-encoded target working condition data is first taken as an input of the first layer of multi-head self-attention network layer, and the input and output of the first layer of multi-head self-attention network layer are fused through a residual connection to serve as an input of the second layer of multi-head self-attention network layer, and the output of the second layer of multi-head self-attention network layer is the output of the Transformer encoder.

[0049] Further, the output of the Transformer encoder is transmitted to an LSTM decoder, referring to Figure 5 , the LSTM decoder introduces a forgetting gate, an input gate and an output gate to build a storage unit, which not only can selectively remember historical information, but also can continuously transmit and update information, effectively alleviating the problem of gradient vanishing or gradient explosion in the training process of traditional regression networks.

[0050] Firstly, the forgetting gate determines the forgetting degree of the input gate structure information, which is calculated according to the input state at the current time and the hidden state at the previous time:

[0051] wherein, represents a Sigmoid function, is an output of the forgetting gate at the current time, and are a weight matrix and a bias matrix of the forgetting gate, respectively. The input state at the current time is the output of the multi-head self-attention mechanism at the current time.

[0052] The input gate determines how much information is retained to the cell state, and the calculation process is as follows:

[0053]

[0054] wherein, represents an adjustment coefficient of the cell state, candidate values of the cell state, denotes a Hyperbolic Tangent function, and is a weight matrix of the input gate, and is a bias matrix of the input gate. Accordingly, the cell state at the current time step can be obtained as :

[0055] Finally, the output gate determines which information is output from the memory cell, and is calculated as follows:

[0056]

[0057] wherein, denotes the memory proportion of the cell state at the current time step, is the output of the output gate at the current time step, and is also the output of the LSTM decoder and the hidden state at the current time step, and are a weight matrix and a bias matrix of the output gate, respectively.

[0058] Finally, the output of the output gate is mapped to the target output dimension by using a ReLu layer, a Dropout layer and a fully connected layer, and the occurrence degree of various fault conditions of the fuel cell is obtained in real time, which can be used as input information for subsequent control decisions.

[0059] Specifically, the hybrid neural network model determines the optimal number of hidden layer nodes, the optimal initial learning rate and the optimal regularization coefficient by using a Bayesian optimization algorithm, is trained by using an adaptive moment estimation optimizer, and uses the mean square error loss as the training loss MSE in the training process:

[0060] wherein, is the number of samples, is the label of the i-th sample, is the predicted value responding to the i-th sample feature.

[0061] During training, all samples can be divided into a training set and a test set based on a fixed random seed at a ratio of 85% and 15%.

[0062] As described above, the fuel cell operation fault diagnosis method in the embodiment is completely described, and the output fault state of the target fuel cell can be applied to the operation management control of the target fuel cell.

[0063] Specifically, refer to Figure 6When the fuel cell system is running, the current load current value and the current impedance spectrum data of the fuel cell (electrical pile in the figure) are obtained through the on-line testing device of the system impedance; then the proposed fuel cell running fault diagnosis method is used to identify the fault state of the fuel cell (including the occurrence degree of membrane dry, water flooding and gas shortage faults), and the fault state is further transmitted to the next level fuel cell running management module as control strategy information reference, and reasonable control information is calculated according to the real-time fault state; further, the running management model outputs the control signal to the fuel cell system, and realizes the impedance-based fuel cell running management control.

[0064] In an exemplary embodiment, a computer device, which can be a server or a terminal, is provided, and an internal structure diagram of the computer device can be as shown in Figure 7 The computer device includes a processor, a memory, an input / output interface (I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through network connection. The computer program is executed by the processor to implement the fuel cell running fault diagnosis method.

[0065] Those skilled in the art can understand that Figure 7 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0066] In an exemplary embodiment, a computer device is also provided, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0067] In an exemplary embodiment, a computer readable storage medium is provided, which stores a computer program. The computer program is executed by a processor to implement the steps in the above method embodiments.

[0068] In an example embodiment, a computer program product is provided, including a computer program which, when executed by a processor, implements the steps of any of the above method embodiments.

[0069] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant regulations.

[0070] It can be understood by those skilled in the art that all or part of the processes in the above-mentioned embodiments can be completed by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiments. Any reference to memory, database or other medium used in the embodiments provided by the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0071] The database involved in the embodiments provided by the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a blockchain, etc., without being limited thereto. The processor involved in the embodiments provided by the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., without being limited thereto.

[0072] Any technical features in the above embodiments can be combined, and for the sake of brevity, not all possible combinations are described above, however, it should be understood that the application encompasses all possible combinations of the technical features unless such a combination is not technically possible.

[0073] The principles and implementation manners of the present application are described herein by using specific examples, and the above embodiments are only used to help understand the method of the present application and its core idea; meanwhile, according to the idea of the present application, the specific implementation manners and application scopes will be changed by those skilled in the art. In conclusion, the content of the present specification should not be understood as a limitation of the present application.

Claims

1. A method of diagnosing a malfunction of an operation of a fuel cell, characterized by, The method comprises: obtaining target working condition data of a target fuel cell in real time, the target working condition data comprising a current load current value and current impedance spectrum data; inputting the target working condition data into a target deep learning model to predict a fault state of the target fuel cell by the target deep learning model, wherein the target deep learning model has learned a mapping relationship between the target working condition data and the fault state through training.

2. The method of diagnosing a malfunction of a fuel cell operation according to claim 1, characterized by, The current impedance spectrum data comprises real part values and imaginary part values of impedance points in a current impedance spectrum. The target working condition data is a data sequence, and elements of the data sequence comprise the current load current value, the real part values and the imaginary part values of the impedance points in the current impedance spectrum.

3. The method of claim 1, wherein The fault state comprises occurrence degrees of various fault conditions, and the various fault conditions comprise membrane dryness, water flooding and gas deficiency.

4. The method of claim 1, wherein The training steps of the target deep learning model comprise: simulating different fault states of a sample fuel cell by adjusting operating conditions of the sample fuel cell; obtaining impedance spectrum data of the sample fuel cell with different load current values in different fault states respectively; training the target deep learning model by taking the load current values and the impedance spectrum data as sample features and taking the fault states as sample labels.

5. The method of claim 4, wherein The operating conditions comprise an air excess coefficient, a water outlet temperature, cathode and anode pressures and relative air humidity.

6. The method of claim 2, wherein The target deep learning model is a hybrid neural network model, which comprises a position encoding layer, a Transformer encoder, an LSTM decoder and a prediction layer, and the Transformer encoder comprises a multi-head self-attention network layer. The position encoding layer is configured to perform position encoding on the target working condition data. The multi-head self-attention network layer is configured to perform feature extraction and fusion on the position-encoded target working condition data to generate an output of the Transformer encoder. The LSTM decoder is configured to decode the output of the Transformer encoder. The prediction layer is configured to predict the fault state according to the output of the LSTM decoder.

7. The method of claim 6, wherein The hybrid neural network model adopts a Bayesian optimization algorithm to determine an optimal number of hidden layer nodes, an optimal initial learning rate and an optimal regularization coefficient, adopts an adaptive moment estimation optimizer for training, and adopts a selected mean square error loss as a training loss during the training process.

8. A computer device comprising: A memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the fuel cell operating fault diagnosis method in any one of claims 1-7.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the fuel cell operating fault diagnosis method in any one of claims 1-7.

10. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the fuel cell operating fault diagnosis method in any one of claims 1-7.

Images