Fuel cell fault diagnosis method and device based on bidirectional long and short time memory network

Abstract

The invention discloses a fuel cell fault diagnosis method and device based on a bidirectional long and short time memory network in the field of fuel cells, and the method comprises the steps: obtaining time step data of a target operation parameter of a fuel cell, and carrying out the preprocessing of the time step data; constructing a bidirectional LSTM-based fault diagnosis model, and training the fault diagnosis model by using the preprocessed time step data; and performing fault diagnosis on the fuel cell by adopting the fault diagnosis model which is finally obtained by training. According to the method, the bidirectional LSTM is utilized to process the time sequence data, the correlation of the data before and after is fully considered, accurate fault detection of each module of the fuel cell system is realized, and the diagnosis precision is improved. Meanwhile, through the lightweight design of the model, the occupation of computing resources is reduced, and the deployment and maintenance cost is reduced, so that the scheme is suitable for a vehicle-mounted controller and other scenes with limited resources.

Description

Technical Field

[0001] The present invention relates to the technical field of fuel cells, and in particular to a fuel cell fault diagnosis method and device based on a bidirectional long short-term memory network. Background Art

[0002] As a clean and efficient energy conversion device, fuel cells have broad application potential in transportation, fixed power stations, portable devices and other fields. Fuel cell systems may be affected by various faults during operation, which may lead to degradation of stack performance, battery aging, and even safety issues. In order to ensure the safety and stability of fuel cell systems, research on fault diagnosis technology has received increasing attention. Fault diagnosis technology can detect faults in the system in a timely manner and take corresponding control measures to avoid more serious consequences.

[0003] Existing fuel cell fault diagnosis methods are mainly divided into three categories: methods based on physical / chemical detection technology, methods based on model-driven, and methods based on data-driven. Physical / chemical detection technology methods directly rely on actual measurements, which are accurate but require a large number of sensors to support them; model-driven methods have advantages in accurately simulating specific fault mechanisms, but they require high accuracy of the model; and data-driven methods can handle complex and multi-dimensional fault modes, have strong adaptability, but rely on a large amount of high-quality data. With the advancement of artificial intelligence and big data technology, data-driven methods are increasingly used in fuel cell fault diagnosis. These methods can identify complex fault modes and trends by analyzing massive historical data and real-time data, further improving the accuracy and real-time performance of diagnosis. In addition, the integration of multiple diagnostic technologies (such as the combination of data-driven and model-driven) has also become a development trend in fault diagnosis. Through the comprehensive analysis of multi-source information, the system status can be judged more comprehensively and accurately.

[0004] For example, a neural network-based fuel cell fault diagnosis method described in patent publication number CN117388702A can provide relatively accurate fault warnings, but its model is based on Transformer. Compared with other neural network models, the training of the Transformer model requires more time and computing resources, and has greater hardware requirements. The model is not suitable for deployment on edge devices with limited computing resources (such as vehicle-mounted controllers), has low diagnostic efficiency, and is not suitable for scenarios with longer real-time requirements. As for the type of fault, the invention only judges two types of water management faults, membrane dryness and flooding, which has certain limitations. Summary of the invention

[0005] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a fuel cell fault diagnosis method and device based on a bidirectional long short-term memory network. Compared with the existing data-driven method, the method has fewer model parameters, shorter training time, stronger real-time reasoning, and is easy to deploy in a vehicle controller with limited computing resources.

[0006] To achieve the above object, the present invention is implemented by adopting the following technical solutions:

[0007] In a first aspect, the present invention provides a fuel cell fault diagnosis method based on a bidirectional long short-term memory network, comprising:

[0008] Acquiring time step data of target operating parameters of the fuel cell and preprocessing the time step data;

[0009] Constructing a fault diagnosis model based on bidirectional LSTM, and training the fault diagnosis model using the preprocessed time step data;

[0010] The fault diagnosis model finally obtained through training is used to perform fault diagnosis on the fuel cell.

[0011] Optionally, the target operating parameters include:

[0012] Stack current , current density , Excess air coefficient , the pressure of hydrogen entering the fuel cell and temperature , Hydrogen pressure out of the stack and temperature , Air pressure into the stack and temperature , Air out of the stack pressure and temperature , the temperature of cooling water in and out of the stack and , minimum cell voltage of the stack , Average voltage of battery stack , mean square error of stack cell voltage , stack impedance and cathode side pressure drop .

[0013] Optionally, the fault diagnosis model includes a bidirectional LSTM module, a first fully connected layer, a PointNet encoder, a second fully connected layer, an average pooling layer, a projection layer, and a classification layer connected in sequence;

[0014] The bidirectional LSTM module performs forward feature processing and reverse feature processing on the input time step data to obtain a bidirectional hidden state :

[0015]

[0016] In the formula, is the time step The hidden state obtained by forward feature processing and reverse feature processing, , is the total number of time steps;

[0017] The first fully connected layer is a bidirectional hidden state Mapping is performed to obtain the feature vector :

[0018]

[0019] In the formula, is the weight matrix and bias of the first fully connected layer;

[0020] The PointNet encoder uses a multi-layer perceptron to encode the feature vector Mapping is performed to obtain the feature vector :

[0021]

[0022] In the formula, is a multi-layer perceptron;

[0023] The second fully connected layer performs a feature vector Mapping is performed to obtain the feature vector :

[0024]

[0025] In the formula, is the weight matrix and bias of the second fully connected layer;

[0026] The average pooling layer performs the feature vector Perform global average pooling to obtain global features :

[0027]

[0028] In the formula, is global average pooling;

[0029] The projection layer is used to predict global features. Mapping to obtain global features :

[0030]

[0031] In the formula, is the weight matrix and bias of the projection layer;

[0032] The classification layer uses the softmax activation function to classify the global features. Mapping is performed to obtain the probability distribution of each fault category and output the final probability distribution of the fault category :

[0033]

[0034] In the formula, is the softmax activation function.

[0035] Optionally, the bidirectional LSTM module is based on the input time step data of the current time step and the hidden state of the previous time step , the forgetting state is calculated through the forget gate, input gate, memory cell and output gate respectively , Input Status , Unit Status And the output status :

[0036]

[0037]

[0038]

[0039]

[0040]

[0041] In the formula, is the sigmoid activation function, for Activation function, is the weight vector, is the bias, is the time step The time step data, is the Hadamard product of the matrix, is an intermediate variable;

[0042] According to the unit status and the output status Calculate the hidden state at the current time step :

[0043] .

[0044] Optionally, the using the preprocessed time step data to train the fault diagnosis model includes:

[0045] Initialize the model parameters of the fault diagnosis model and repeat the following steps until the model parameters converge or the preset number of iterations is reached:

[0046] Inputting the time step data into the fault diagnosis model to obtain the probability distribution of the corresponding fault category;

[0047] Calculate the loss using a cross entropy loss function based on the probability distribution of the fault class and the true label of the fault class;

[0048] The gradient of the loss to the model parameters is calculated by the back-propagation algorithm, and the model parameters are updated according to the gradient by the Adam optimizer.

[0049] Optionally, the cross entropy loss function for:

[0050]

[0051] In the formula, Fault category output by the fault diagnosis model The probability distribution of is the total number of fault categories, Fault Category The true label value of .

[0052] In a second aspect, the present invention provides a fuel cell fault diagnosis device based on a bidirectional long short-term memory network, comprising:

[0053] A data acquisition processing module is configured to acquire time step data of a target operating parameter of the fuel cell and preprocess the time step data;

[0054] A model building training module is configured to build a fault diagnosis model based on a bidirectional LSTM and train the fault diagnosis model using the preprocessed time step data;

[0055] The battery fault diagnosis module is configured to perform fault diagnosis on the fuel cell using the fault diagnosis model finally obtained through training.

[0056] In a third aspect, the present invention provides an electronic device, including a processor and a storage medium;

[0057] The storage medium is used to store instructions;

[0058] The processor is used to operate according to the instructions to execute the steps according to the above method.

[0059] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the above method when executed by a processor.

[0060] In a fifth aspect, the present invention provides a computer program product, comprising a computer program / instruction, which implements the steps of the above method when executed by a processor.

[0061] Compared with the prior art, the present invention has the following beneficial effects:

[0062] The present invention provides a fuel cell fault diagnosis method and device based on a bidirectional long short-term memory network, which can perform efficient and accurate fault detection: using bidirectional LSTM to process time series data, fully considering the before and after correlation of data, to achieve accurate fault detection of each module of the fuel cell system (hydrogen supply, air supply, water management), and improve the diagnostic accuracy. It can identify multiple fault types: it can identify common fault states of multiple fuel cells, such as membrane drying, water flooding, hydrogen hunger, oxygen hunger, etc., to provide comprehensive information for system maintenance and improve the reliability of the system; it can monitor and warn in real time: through bidirectional LSTM to perform real-time data analysis on the fuel cell system, it can timely discover abnormal states and trigger warnings, effectively avoid major faults, and extend the operating life of the system; it has a lightweight design: through the lightweight design of the model, the occupation of computing resources is reduced, and the deployment and maintenance costs are reduced, so that the scheme is suitable for scenarios with limited resources such as vehicle-mounted controllers. In summary, the present invention can significantly improve the fault detection capability and operating efficiency of the fuel cell system, and promote the popularization of fuel cell vehicles and other fuel cell applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 It is a flowchart of a fuel cell fault diagnosis method based on a bidirectional long short-term memory network provided by an embodiment of the present invention;

[0064] Figure 2 is a structural diagram of a fault diagnosis model provided by an embodiment of the present invention;

[0065] Figure 3 It is a schematic diagram of a process for training a fault diagnosis model provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0066] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0067] Embodiment 1:

[0068] like Figure 1As shown, an embodiment of the present invention provides a fuel cell fault diagnosis method based on a bidirectional long short-term memory network, comprising the following steps:

[0069] Step S1, obtaining time step data of target operating parameters of the fuel cell, and preprocessing the time step data.

[0070] Specifically in this embodiment, in order to efficiently and accurately diagnose and warn of faults in the hydrogen supply module, air supply module and water management module of the fuel cell system, the selection of target operating parameters includes:

[0071] Stack current , current density , Excess air coefficient , the pressure of hydrogen entering the fuel cell and temperature , Hydrogen pressure out of the stack and temperature , the pressure of air entering the stack and temperature , Air out of the stack pressure and temperature , the temperature of cooling water in and out of the stack and , minimum cell voltage of the stack , Average voltage of battery stack , mean square error of stack cell voltage , stack impedance and cathode side pressure drop .

[0072] In other optional implementations, the staff can adjust and select other operating parameters as needed, which will not be described in detail in this implementation.

[0073] Preprocessing of time step data generally includes data missing value processing and normalization processing. Missing value processing processes the missing values ​​in the data to ensure the integrity and accuracy of the data. General methods include: deletion, filling, interpolation, etc. Normalization processing converts data of different magnitudes or distributions to the same magnitude or distribution range for subsequent data analysis and model training.

[0074] Step S2: construct a fault diagnosis model based on bidirectional LSTM, and use the preprocessed time step data to train the fault diagnosis model.

[0075] Bidirectional LSTM (BiLSTM for short) is a special recurrent neural network (RNN) architecture that extends the capabilities of RNN and can use both previous and subsequent information in the sequence for prediction. Bidirectional LSTM consists of an input layer, a hidden layer, and an output layer. The input layer takes the input state sequence as input, and the hidden layer contains recurrent connections that allow the network to maintain memory of past states. The output layer produces predicted future states. Compared with LSTM, bidirectional LSTM plays a significant role in fault warning, especially in the analysis and prediction of time series data. By capturing the contextual information of the sequence from both the forward and backward directions, it can better detect potential fault trends in the system, thereby improving the accuracy and response speed of fault warning. In this embodiment, a fault diagnosis model is constructed based on bidirectional LSTM, which can improve the accuracy of fault diagnosis.

[0076] like Figure 2 As shown, specifically in this embodiment, the fault diagnosis model includes a bidirectional LSTM module, a first fully connected layer, a PointNet encoder, a second fully connected layer, an average pooling layer, a projection layer and a classification layer connected in sequence.

[0077] (1) The bidirectional LSTM module performs forward feature processing and reverse feature processing on the input time step data to obtain a bidirectional hidden state :

[0078]

[0079] In the formula, is the time step The hidden state obtained by forward feature processing and reverse feature processing, , is the total number of time steps;

[0080] The bidirectional LSTM module is based on the input time step data of the current time step and the hidden state of the previous time step. , the forgetting state is calculated through the forget gate, input gate, memory cell and output gate respectively , Input Status , Unit Status And the output status :

[0081]

[0082]

[0083]

[0084]

[0085]

[0086] In the formula, is the sigmoid activation function, for Activation function, is the weight vector, is the bias, is the time step The time step data, is the Hadamard product of the matrix, is an intermediate variable;

[0087] According to the unit status and output status Calculate the hidden state at the current time step :

[0088] .

[0089] (2) The first fully connected layer has two-way hidden states Mapping is performed to obtain the feature vector :

[0090]

[0091] In the formula, is the weight matrix and bias of the first fully connected layer.

[0092] (3) PointNet encoder uses a multi-layer perceptron to encode feature vectors Mapping is performed to obtain the feature vector :

[0093]

[0094] In the formula, It is a multi-layer perceptron.

[0095] (4) The second fully connected layer performs feature vector Mapping is performed to obtain the feature vector :

[0096]

[0097] In the formula, is the weight matrix and bias of the second fully connected layer.

[0098] (5) Average pooling layer for feature vector Perform global average pooling to obtain global features :

[0099]

[0100] In the formula, is global average pooling.

[0101] (6) Projection layer for global features Mapping to obtain global features :

[0102]

[0103] In the formula, is the weight matrix and bias of the projection layer.

[0104] (7) The classification layer uses the softmax activation function to optimize the global features Mapping is performed to obtain the probability distribution of each fault category and output the final probability distribution of the fault category :

[0105]

[0106] In the formula, is the softmax activation function.

[0107] This implementation combines bidirectional LSTM and PointNet to effectively extract local and global features in time series. Bidirectional LSTM processes the dependencies between the previous and next steps in the time series, while PointNet is responsible for capturing the overall features between different time steps. This approach can more accurately identify faults in fuel cell systems and diagnose and warn of faults.

[0108] like Figure 3 As shown, specifically in this embodiment, using the preprocessed time step data to train the fault diagnosis model includes:

[0109] Step S01, initialize the model parameters of the fault diagnosis model, and repeat steps S02 to S04 until the model parameters converge or reach a preset number of iterations:

[0110] Step S02: input the time step data into the fault diagnosis model to obtain the probability distribution of the corresponding fault category;

[0111] Step S03, calculating the loss using a cross entropy loss function according to the probability distribution of the fault category and the true label of the fault category;

[0112] Cross Entropy Loss Function for:

[0113]

[0114] In the formula, Fault category output by the fault diagnosis model The probability distribution of is the total number of fault categories, Fault Category The true label value of .

[0115] Specifically in this embodiment, the present invention uses the target operating parameters of the hydrogen supply module, the air supply module and the water management module to diagnose four fault states (membrane dryness, flooding, hydrogen starvation, oxygen starvation) and the normal state of the fuel cell, so The value is 5.

[0116] Step S04: Calculate the gradient of the loss to the model parameters through the back propagation algorithm, and update the model parameters according to the gradient through the Adam optimizer;

[0117]

[0118]

[0119] In the formula, are model parameters, is the learning rate, For loss.

[0120] Step S3: using the fault diagnosis model finally obtained through training to perform fault diagnosis on the fuel cell.

[0121] In summary, the fuel cell fault diagnosis method based on the bidirectional long short-term memory network provided by the present invention collects relevant sensor data and controller data from the fuel cell system, screens the collected raw data, and ensures that the characteristic data is suitable for input into the model for fault detection. The preprocessed data is analyzed using a trained bidirectional LSTM neural network model to identify potential faults of the system. According to the results output by the model, it is determined whether the system is normal or a fault is detected. If the model determines that the system is normal, the status information of "system normal" is output. If the model detects a fault, the specific fault type is output, including membrane dryness, water flooding, hydrogen hunger, and oxygen hunger.

[0122] Embodiment 2:

[0123] The embodiment of the present invention provides a fuel cell fault diagnosis device based on a bidirectional long short-term memory network, comprising:

[0124] A data acquisition processing module is configured to acquire time step data of target operating parameters of the fuel cell and pre-process the time step data;

[0125] A model building and training module is configured to build a fault diagnosis model based on a bidirectional LSTM and train the fault diagnosis model using the preprocessed time step data;

[0126] The battery fault diagnosis module is configured to perform fault diagnosis on the fuel cell using the fault diagnosis model finally obtained through training.

[0127] Embodiment three:

[0128] An embodiment of the present invention provides an electronic device, including a processor and a storage medium;

[0129] The storage medium is used to store instructions;

[0130] The processor is used to operate according to the instructions to execute the steps according to the above method.

[0131] Embodiment 4:

[0132] An embodiment of the present invention provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the above method are implemented.

[0133] Embodiment five:

[0134] An embodiment of the present invention provides a computer program product, including a computer program / instruction, which implements the steps of the above method when executed by a processor.

[0135] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0136] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0137] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0138] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0139] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A fuel cell fault diagnosis method based on a bidirectional long short-term memory network, characterized in that: include: Acquiring time step data of target operating parameters of the fuel cell and preprocessing the time step data; Constructing a fault diagnosis model based on bidirectional LSTM, and training the fault diagnosis model using the preprocessed time step data; The fault diagnosis model finally obtained through training is used to perform fault diagnosis on the fuel cell.

2. The fuel cell fault diagnosis method based on bidirectional long short-term memory network according to claim 1 is characterized in that: The target operating parameters include: Stack current , current density , Excess air coefficient , the pressure of hydrogen entering the fuel cell and temperature , Hydrogen pressure out of the stack and temperature , the pressure of air entering the stack and temperature , Air out of the stack pressure and temperature , the temperature of cooling water in and out of the stack and , minimum cell voltage of the stack , Average voltage of stack cells , mean square error of stack cell voltage , stack impedance and cathode side pressure drop .

3. The fuel cell fault diagnosis method based on bidirectional long short-term memory network according to claim 1 is characterized in that: The fault diagnosis model includes a bidirectional LSTM module, a first fully connected layer, a PointNet encoder, a second fully connected layer, an average pooling layer, a projection layer and a classification layer connected in sequence; The bidirectional LSTM module performs forward feature processing and reverse feature processing on the input time step data to obtain a bidirectional hidden state : In the formula, is the time step The hidden state obtained by forward feature processing and reverse feature processing, , is the total number of time steps; The first fully connected layer is a bidirectional hidden state Mapping is performed to obtain the feature vector : In the formula, is the weight matrix and bias of the first fully connected layer; The PointNet encoder uses a multi-layer perceptron to encode the feature vector Mapping is performed to obtain the feature vector : In the formula, is a multi-layer perceptron; The second fully connected layer performs a feature vector Mapping is performed to obtain the feature vector : In the formula, is the weight matrix and bias of the second fully connected layer; The average pooling layer performs the feature vector Perform global average pooling to obtain global features : In the formula, is global average pooling; The projection layer is used to predict global features. Mapping to obtain global features : In the formula, is the weight matrix and bias of the projection layer; The classification layer uses the softmax activation function to classify the global features. Mapping is performed to obtain the probability distribution of each fault category and output the final probability distribution of the fault category : In the formula, is the softmax activation function.

4. The fuel cell fault diagnosis method based on bidirectional long short-term memory network according to claim 3 is characterized in that: The bidirectional LSTM module is based on the input time step data of the current time step and the hidden state of the previous time step , the forgetting state is calculated through the forget gate, input gate, memory cell and output gate respectively , Input Status , Unit Status And the output status : In the formula, is the sigmoid activation function, for Activation function, is the weight vector, is the bias, is the time step The time step data, is the Hadamard product of the matrix, is an intermediate variable; According to the unit status and the output status Calculate the hidden state at the current time step : 。 5. The fuel cell fault diagnosis method based on bidirectional long short-term memory network according to claim 3 is characterized in that: The using the preprocessed time step data to train the fault diagnosis model comprises: Initialize the model parameters of the fault diagnosis model and repeat the following steps until the model parameters converge or the preset number of iterations is reached: Inputting the time step data into the fault diagnosis model to obtain the probability distribution of the corresponding fault category; Calculate the loss using a cross entropy loss function based on the probability distribution of the fault class and the true label of the fault class; The gradient of the loss to the model parameters is calculated by the back-propagation algorithm, and the model parameters are updated according to the gradient by the Adam optimizer.

6. The fuel cell fault diagnosis method based on bidirectional long short-term memory network according to claim 5 is characterized in that: The cross entropy loss function for: In the formula, Fault category output by the fault diagnosis model The probability distribution of is the total number of fault categories, Fault Category The true label value of .

7. A fuel cell fault diagnosis device based on a bidirectional long short-term memory network, characterized in that: include: A data acquisition processing module is configured to acquire time step data of a target operating parameter of the fuel cell and preprocess the time step data; A model building training module is configured to build a fault diagnosis model based on a bidirectional LSTM and train the fault diagnosis model using the preprocessed time step data; The battery fault diagnosis module is configured to perform fault diagnosis on the fuel cell using the fault diagnosis model finally obtained through training.

8. An electronic device, characterized in that: including processor and storage medium; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the method according to any one of claims 1-5.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images