Battery fault diagnosis model training and battery fault diagnosis method, equipment and medium

Abstract

The invention discloses a battery fault diagnosis model training and battery fault diagnosis method, equipment and a medium. The method comprises the following steps: acquiring a membrane electrode assembly voltage signal of a proton exchange membrane fuel cell and a fault label value corresponding to the membrane electrode assembly voltage signal; performing variational mode decomposition and integrated empirical mode decomposition on the membrane electrode assembly voltage signal to obtain a sample feature vector corresponding to the membrane electrode assembly voltage signal; an initial battery fault diagnosis model is trained based on the sample feature vectors and the fault label values to obtain a target battery fault diagnosis model, and the initial battery fault diagnosis model comprises a multi-scale time convolution module, a mixed attention mechanism module and a Ghost module. Through the target battery fault diagnosis model, the accuracy of battery fault diagnosis can be improved, and various fault types of the hydrogen fuel battery can be effectively identified.

Description

Technical Field

[0001] The present invention relates to the technical field of fuel cell fault diagnosis, and in particular to battery fault diagnosis model training and battery fault diagnosis methods, equipment and media. Background Art

[0002] The proton exchange membrane fuel cell (PEMFC) system is an extremely complex multi-physics coupling system, including various components such as air supply system, temperature control module, humidification tank, fuel cell stack, etc. On the one hand, it is difficult to accurately establish a PEMFC system model to reliably monitor and manage its health status. On the other hand, a large amount of data that can reflect its operating status is generated at all times during the operation of PEMFC, and the robustness and reliability of fault diagnosis methods based on signal analysis will face huge challenges when facing complex massive data. Therefore, based on cost optimization, practical application limitations, technical business opportunities and other factors, how to reasonably and efficiently develop and utilize these massive data to ensure the reliability of PEMFC system operation has become an urgent problem to be solved.

[0003] When in the same PEMFC health state, the operating data have certain common characteristics to a certain extent, which are reflected in the characteristic information hidden in the data. However, the fault distinguishability of the time domain, frequency domain or shallow characteristics in the PEMFC voltage signal is not prominent, which makes the fault diagnosis accuracy of the current related algorithms not high. Summary of the invention

[0004] The present invention provides a battery fault diagnosis model training and battery fault diagnosis method, device and medium to improve the accuracy of battery fault diagnosis and effectively identify various fault types of hydrogen fuel cells.

[0005] According to one aspect of the present invention, a battery fault diagnosis model training method is provided, comprising:

[0006] Acquire a membrane electrode assembly voltage signal of a proton exchange membrane fuel cell and a fault label value corresponding to the membrane electrode assembly voltage signal;

[0007] Performing variational mode decomposition and integrated empirical mode decomposition on the membrane electrode assembly voltage signal to obtain a sample feature vector corresponding to the membrane electrode assembly voltage signal;

[0008] An initial battery fault diagnosis model is trained based on the sample feature vector and the fault label value to obtain a target battery fault diagnosis model, wherein the initial battery fault diagnosis model includes a multi-scale time convolution module, a hybrid attention mechanism module and a Ghost module.

[0009] According to another aspect of the present invention, a battery fault diagnosis method is provided, comprising:

[0010] Acquiring a voltage signal of a membrane electrode assembly to be diagnosed of a battery to be diagnosed;

[0011] Processing the voltage signal of the membrane electrode assembly to be diagnosed by variational mode decomposition and integrated empirical mode decomposition to obtain a characteristic vector to be diagnosed of the battery to be diagnosed;

[0012] The feature vector to be diagnosed is input into the target battery fault diagnosis model to obtain a target diagnosis result corresponding to the battery to be diagnosed output by the target battery fault diagnosis model.

[0013] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0014] at least one processor;

[0015] and a memory communicatively connected to the at least one processor; wherein,

[0016] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the battery fault diagnosis model training method or the battery fault diagnosis method described in any embodiment of the present invention.

[0017] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the battery fault diagnosis model training method or battery fault diagnosis method described in any embodiment of the present invention when executed.

[0018] The technical solution of the embodiment of the present invention obtains the membrane electrode assembly voltage signal of the proton exchange membrane fuel cell and the fault label value corresponding to the membrane electrode assembly voltage signal; performs variational mode decomposition and integrated empirical mode decomposition on the membrane electrode assembly voltage signal to obtain the sample feature vector corresponding to the membrane electrode assembly voltage signal, which can effectively mine the feature information hidden in the data and solve the problem that the fault distinguishability of the time domain, frequency domain or shallow features in the membrane electrode assembly voltage signal is not prominent, resulting in the low fault diagnosis accuracy of the current related algorithms. The sample feature vector is further used to train and classify the neural network model to establish a target battery fault diagnosis model, which can improve the fault diagnosis accuracy and effectively identify various fault types of hydrogen fuel cells.

[0019] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0021] Figure 1 A process of a battery fault diagnosis model training method provided in Embodiment 1 of the present invention;

[0022] Figure 2 A flow chart of feature extraction of a membrane electrode assembly voltage signal provided in the first embodiment of the present invention;

[0023] Figure 3 A flowchart of a battery fault diagnosis model training method provided in Embodiment 2 of the present invention;

[0024] Figure 4 A structural diagram of a multi-scale temporal convolution module provided in Embodiment 2 of the present invention;

[0025] Figure 5 A structural diagram of a hybrid attention mechanism module provided in Embodiment 2 of the present invention;

[0026] Figure 6 A structural diagram of the Ghost 1D module provided in Embodiment 2 of the present invention;

[0027] Figure 7 The overall structure diagram of the target battery fault diagnosis model provided in the second embodiment of the present invention;

[0028] Figure 8 The loss curve during the training process of the target battery fault diagnosis model;

[0029] Fig. 9 The accuracy change curve of the target battery fault diagnosis model during training;

[0030] Fig.10 A flowchart of a battery fault diagnosis method provided in Embodiment 3 of the present invention;

[0031] Fig.11 A schematic diagram of the structure of a battery fault diagnosis model training device provided in Embodiment 4 of the present invention;

[0032] Fig.12This is a schematic diagram of the structure of an electronic device provided in Embodiment 5 of the present invention. DETAILED DESCRIPTION

[0033] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0034] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0035] Embodiment 1

[0036] Figure 1 This is a flow chart of a battery fault diagnosis model training method provided in the first embodiment of the present invention. This embodiment is applicable to the case of fault diagnosis of proton exchange membrane fuel cells. The method can be executed by a battery fault diagnosis model training device, which can be implemented in the form of hardware and / or software, and can be configured in a computer device. Figure 1 As shown, the method includes:

[0037] S110, obtaining a membrane electrode assembly voltage signal of a proton exchange membrane fuel cell and a fault label value corresponding to the membrane electrode assembly voltage signal.

[0038] Among them, the membrane electrode assembly (MEA) voltage signal; the fault label value can be a pre-set fault type label, which is used to identify the fault type corresponding to the current membrane electrode assembly voltage signal of the proton exchange membrane fuel cell.

[0039] In an embodiment of the present invention, the method of obtaining a membrane electrode assembly voltage signal of a proton exchange membrane fuel cell includes: placing the proton exchange membrane fuel cell in various operating conditions through a proton exchange membrane fuel cell test platform; obtaining a to-be-processed voltage signal of the membrane electrode assembly of the proton exchange membrane fuel cell under various operating conditions, and normalizing the to-be-processed voltage signal to obtain the membrane electrode assembly voltage signal.

[0040] The type of the operating condition includes at least one of a normal operating condition, a flooding fault, a dehydration fault, an air starvation fault, and a hydrogen starvation fault. The voltage signal to be processed refers to a voltage signal that has not been pre-processed.

[0041] Specifically, the parameters of the proton exchange membrane fuel cell test platform can be set so that the proton exchange membrane fuel cell is in different working conditions in turn, and then the voltage signals to be processed of the membrane electrode assembly under different working conditions are collected. Then, the voltage signals to be processed are normalized to obtain the membrane electrode assembly voltage signals.

[0042] Correspondingly, the fault label value can be the fault type. For example, the parameters of the proton exchange membrane fuel cell test platform are set so that the proton exchange membrane fuel cell is in a water flooding fault condition. At this time, the membrane electrode assembly voltage signal of the proton exchange membrane fuel cell can be collected, and the water flooding fault is used as the fault label value.

[0043] In an optional implementation, in order to make the voltage of PEMFC conform to the normal distribution and suppress the negative impact caused by abnormal data values, thereby reducing the difficulty of network training and accelerating the convergence speed, Z-score normalization is used to process the voltage signal to be processed. The mathematical expression of Z-score normalization is as follows:

[0044]

[0045] Among them, x * The data is normalized, mean means the mean, and std means the standard deviation.

[0046] In a preferred embodiment, the parameters of the proton exchange membrane fuel cell test platform are set to put the proton exchange membrane fuel cell in different working conditions, which can be different hydrogen stoichiometric ratios, air stoichiometric ratios, humidification tank temperatures, and whether to add humidified gas, etc., to simulate the conditions of various working conditions. The following are the parameter settings under various working conditions, among which the battery temperature is set to 50°C and the current density is set to 0.6A / cm 2 :

[0047] Normal operation: hydrogen stoichiometric ratio is 1.5, air stoichiometric ratio is 2.5, humidification tank temperature is 44°C, and humidification gas is added;

[0048] Water flooding fault: hydrogen stoichiometric ratio is 1.5, air stoichiometric ratio is 2.5, humidification tank temperature is 60°C, and humidification gas is added; (By increasing the humidification tank temperature, when the gas with a higher dew point enters the internal environment of the low-temperature battery, the water vapor in the gas will cool and condense into a large amount of liquid water, causing excessive water accumulation inside the MEA, blocking the gas channel, affecting the transmission of oxygen and hydrogen, and causing water flooding)

[0049] Dehydration failure: hydrogen stoichiometric ratio is 1.5, air stoichiometric ratio is 2.5, humidification tank temperature is 44°C, and no humidification gas is added; (unhumidified dry gas is used as the gas source (relative humidity 0%), so the only source of liquid water inside the battery is the reaction product water, which reduces the water content of the proton exchange membrane, further affecting the proton transmission capacity of the proton exchange membrane and increasing polarization loss)

[0050] Air starvation fault: hydrogen stoichiometric ratio is 1.5, air stoichiometric ratio is 1.5, humidification tank temperature is 44°C, and humidification gas is added; (When the battery cathode is not supplied with enough air, not only will the cathode end undergo an oxygen reduction reaction, but it will also induce a hydrogen evolution reaction, namely the hydrogen pump effect, which will cause the cathode potential to decrease and the output voltage to decrease)

[0051] Hydrogen starvation failure: hydrogen stoichiometric ratio is 1.15, air stoichiometric ratio is 3.5, humidification tank temperature is 44°C, and humidified gas is added; (By reducing the hydrogen stoichiometric ratio from 1.5 to 1.15, when hydrogen starvation occurs, the corrosion rate of the carbon carrier will be accelerated, thereby reducing the electrochemical active surface area of ​​the PEMFC, and the battery voltage will be greatly reduced or even negative)

[0052] According to the above parameter settings, a large number of simulations are performed to collect the voltage signals in the membrane electrode assembly (MEA) and construct an original sample library for subsequent model training. That is, in order to obtain a large number of samples for subsequent training of the initial battery fault diagnosis model, the proton exchange membrane fuel cell test platform can be used to simulate different working conditions multiple times to obtain multiple sets of membrane electrode assembly voltage signals and corresponding fault label values ​​as samples in the original sample library.

[0053] S120, performing variational mode decomposition and integrated empirical mode decomposition on the membrane electrode assembly voltage signal to obtain a sample feature vector corresponding to the membrane electrode assembly voltage signal.

[0054] The sample feature vector may be a feature vector obtained by decomposing the membrane electrode assembly voltage signal.

[0055] In an embodiment of the present invention, variational mode decomposition and integrated empirical mode decomposition are performed on the membrane electrode assembly voltage signal to obtain a sample feature vector corresponding to the membrane electrode assembly voltage signal, including: decomposing the membrane electrode assembly voltage signal into multiple inherent mode components by variational mode decomposition; decomposing the membrane electrode assembly voltage signal into multiple inherent mode components by integrated empirical mode decomposition; determining the energy value of each inherent mode component, and constructing the sample feature vector based on the energy value of each inherent mode component.

[0056] Specifically, the embodiment of the present invention can combine variational mode decomposition (VMD) and ensemble empirical mode decomposition (EEMD) to decompose the MEA voltage signal, and obtain multiple intrinsic mode components (IMF) corresponding to the low-frequency and high-frequency regions of the signal, respectively, and obtain the energy of each IMF component to construct a feature vector, that is, a sample feature vector.

[0057] It should also be noted that the above is an introduction to the process of determining the sample feature vector corresponding to one set of membrane electrode assembly voltage signals, and other membrane electrode assembly voltage signals in the original sample library can be converted into sample feature vectors according to the above process.

[0058] In a preferred embodiment, the feature extraction process of the membrane electrode assembly voltage signal is described in detail, such as Figure 2 As shown, it is a flow chart of feature extraction of membrane electrode assembly voltage signal provided in the first embodiment of the present invention. Figure 2 The feature extraction process using VMD decomposition on the left is as follows:

[0059] Variational mode decomposition decomposes the original signal into K IMFs. Let the sum of all IMF components equal to the original signal, so that the estimated bandwidth of all modal IMFs is minimized. The mathematical model is shown in formula (2).

[0060]

[0061] Where: a is the original signal, u k is the kth modal component of a, ω k for u k The center frequency, is the Hilbert-Huang transform of the kth IMF, and δ(t) is the Dirac distribution function.

[0062] The main iteration process of VMD is:

[0063] (1) Initialization and the maximum number of iterations N, let n = 0;

[0064] (2) Update using formula (3) (4)

[0065]

[0066]

[0067] Wherein, λ is the Lagrange multiplier, and λ(ω) is the frequency domain representation of the Lagrange multiplier λ(t).

[0068] (3) Update using formula (5)

[0069]

[0070] Where: is the Fourier transform of f(ω), f(ω) is the center frequency of the modal component, which indicates the position of the modal component on the spectrum.

[0071] (4) Accuracy convergence criterion ε>0, if not satisfied And n<N, then repeat step 2 until the iteration ends. Output the final

[0072] (5) Through the inverse Fourier transform method, using Get K IMFs.

[0073] After using VMD to decompose multiple IMF components, the energy value of each component is calculated. The square of the absolute value of each value is defined to represent the energy contained in the value. The energy of the IMF component is the sum of the squares of the absolute values ​​of each value, as shown in the following formula (6):

[0074]

[0075] Where N is the number of IMF components; IMF k is the kth IMF component;

[0076] Figure 2 The right side is the feature extraction process using EEMD decomposition, as follows:

[0077] Step 1: Copy the original signal x(t) n times to form the original signal sequence {x i (t), (i=1,2,...,n)}, add Gaussian white noise n to the sequence i (t):

[0078] x i (t) = x(t) + ni (t) (7)

[0079] Step 2: Perform EMD decomposition on the original signal sequence to obtain the two parts in formula (8), c ij represents the IMF component, r i Represents the remainder of a signal:

[0080]

[0081] Step 3: The IMF components decomposed in the second step are matched one by one, and the overall average calculation is performed, and finally the n intrinsic mode components c of the original signal are obtained. j (t) and remainder r(t):

[0082]

[0083] in

[0084] Then, the energy calculation method shown in formula (6) is used again to calculate the energy of each IMF component obtained by EEMD decomposition. The energies obtained by the two decomposition algorithms are combined to construct the feature vector as shown in formula (10):

[0085] T=[E 1,1 ,E 1,2 ,…E 1,N ,E 2,1 ,E 2,2 ,…E 2,n ] (10)

[0086] Where: E 1,1 is the energy of IMF1 obtained by VMD decomposition, E 1,N IMF is obtained by VMD decomposition N The energy, E 2,1 is the energy of IMF1 obtained by EEMD decomposition, E 2,n IMF is obtained by EEMD decomposition n energy.

[0087] S130: training an initial battery fault diagnosis model based on the sample feature vector and the fault label value to obtain a target battery fault diagnosis model.

[0088] Among them, the initial battery fault diagnosis model includes a multi-scale time convolution module, a hybrid attention mechanism module and a Ghost module.

[0089] Specifically, the sample feature vector can be input into the initial battery fault diagnosis model, and the initial battery fault diagnosis model can output the diagnosis prediction result corresponding to the sample feature vector. Based on the diagnosis prediction result and the fault label, the corresponding loss value is calculated, and the model parameters of the initial battery fault diagnosis model are corrected according to the loss value. The process is repeated multiple times to finally obtain a trained target battery fault diagnosis model.

[0090] The technical solution of the embodiment of the present invention obtains the membrane electrode assembly voltage signal of the proton exchange membrane fuel cell and the fault label value corresponding to the membrane electrode assembly voltage signal; performs variational mode decomposition and integrated empirical mode decomposition on the membrane electrode assembly voltage signal to obtain the sample feature vector corresponding to the membrane electrode assembly voltage signal, which can effectively mine the feature information hidden in the data and solve the problem that the fault distinguishability of the time domain, frequency domain or shallow features in the membrane electrode assembly voltage signal is not prominent, resulting in the low fault diagnosis accuracy of the current related algorithms. The sample feature vector is further used to train and classify the neural network model to establish a target battery fault diagnosis model, which can improve the fault diagnosis accuracy and effectively identify various fault types of hydrogen fuel cells.

[0091] Embodiment 2

[0092] Figure 3 This is a flowchart of a battery fault diagnosis model training method provided in the second embodiment of the present invention. Based on the above embodiment, this embodiment further optimizes the battery fault diagnosis model training process. The specific implementation method can refer to the technical solution of this embodiment. Among them, the technical terms that are the same or corresponding to the above embodiment are not repeated here. Figure 3 As shown, the method includes:

[0093] In an embodiment of the present invention, an algorithm operating environment can be built on a deep learning workstation and a neural network can be constructed. Based on the Pycharm software environment, Python3.7 and pytorch deep learning framework, a neural network classification model, i.e., an initial battery fault diagnosis model, is established. The model uses three new network structures, namely, a hybrid attention mechanism module, a multi-scale time convolution module, and a Ghost1D module.

[0094] S210 , obtaining a membrane electrode assembly voltage signal of a proton exchange membrane fuel cell and a fault label value corresponding to the membrane electrode assembly voltage signal.

[0095] S220, performing variational mode decomposition and integrated empirical mode decomposition on the membrane electrode assembly voltage signal to obtain a sample feature vector corresponding to the membrane electrode assembly voltage signal.

[0096] S230: Input the sample feature vector into the multi-scale time convolution module to obtain a first feature vector output by the multi-scale time convolution module.

[0097] The first feature vector may be a feature vector output by a multi-scale temporal convolution module.

[0098] In an embodiment of the present invention, the multi-scale temporal convolution module includes three 1×1 convolution layers, two 3x3 convolution layers and one 5x5 convolution layer. Figure 4 , which is a structural diagram of a multi-scale time convolution module provided in Embodiment 2 of the present invention.

[0099] In an embodiment of the present invention, the sample feature vector is input into the multi-scale time convolution module to obtain a first feature vector output by the multi-scale time convolution module, including: reducing the dimension of the sample feature vector through two 1×1 convolution layers respectively to obtain a first reduced dimension feature vector and a second reduced dimension feature vector; performing time feature information on the first reduced dimension feature vector through two 3x3 convolution layers, one 5x5 convolution layer and one 1×1 convolution layer to obtain a time feature vector; and concatenating the time feature vector with the second reduced dimension feature vector to obtain the first feature vector.

[0100] Specifically, the multi-scale temporal convolution module first uses two 1×1 convolutions to reduce the input features to half of the original dimension, that is, to obtain the first reduced dimension feature vector and the second reduced dimension feature vector. Then, 1×1, 3×3, and 5×5 convolutions are used to extract the temporal feature information of the left branch, and the multi-scale information is concatenated using the concat operation. The 3×3 convolution kernel further mines the deep features to obtain the temporal feature vector. Finally, the two branch features are concatenated as the output, that is, the temporal feature vector is concatenated with the second reduced dimension feature vector to obtain the first feature vector.

[0101] S240. Input the first feature vector into the hybrid attention mechanism module to obtain a second feature vector corresponding to the first feature vector.

[0102] The hybrid attention mechanism module includes a spatial attention mechanism (Coordinate Attention, CA attention mechanism) and a squeeze-and-excitation attention (Squeeze-and-Excitation Attention, SE attention mechanism), and the second feature vector can be a feature vector output by the hybrid attention mechanism module. Figure 5 As shown, it is a structural diagram of the hybrid attention mechanism module provided in Example 2 of the present invention.

[0103] In an embodiment of the present invention, the first feature vector is input into the hybrid attention mechanism module to obtain a second feature vector corresponding to the first feature vector, including: inputting the first feature vector into the spatial attention to obtain a spatial attention feature vector; inputting the spatial attention feature vector into the squeeze and incentive attention to obtain a squeeze and incentive attention feature vector; splicing the spatial attention feature vector and the squeeze and incentive attention feature vector slices to obtain the second feature vector.

[0104] Specifically, the input signal is weighted for each dimensional channel through the CA attention mechanism, and weights are assigned to different channels. Then, three convolutional layers of different sizes are set to extract features of different scales. The signals of different scales are concatenated using Concat, and then processed again using the SE attention mechanism and combined with the data processed by the CA attention mechanism for output.

[0105] S250: Input the second feature vector to the Ghost module to obtain a feature map to be used output by the Ghost module.

[0106] like Figure 6 As shown, it is a structural diagram of the Ghost1D module provided in the second embodiment of the present invention.

[0107] Specifically, the Ghost module first generates m feature maps by using ordinary convolution, then obtains n Ghost feature maps by grouped linear operations, and finally concatenates the feature maps to output m×n feature maps. Since the traditional convolution parameter is A1 and the Ghost convolution parameter is A2, the comparison of the two parameters is calculated by formula (11). It can be seen that the Ghost convolution parameter is approximately 1 / n of the traditional convolution calculation amount.

[0108]

[0109] Where m is the number of feature maps generated by ordinary convolution; n is the number of feature maps obtained by grouped linear operation; A1 is the traditional convolution parameter; A2 is the Ghost convolution parameter; k is the size of the ordinary convolution kernel; d is the spatial dimension of the input data; c is the number of channels of the input data.

[0110] Since the Ghost module is usually used to extract two-dimensional image features, and the PEMFC fault features are one-dimensional features, it is necessary to replace the two-dimensional convolution in the Ghost module with a one-dimensional convolution, and to construct a Ghost convolution that can be applied to one-dimensional signals while keeping the rest of the process unchanged.

[0111] S260 , processing the feature map to be used through a fully connected layer to obtain a fault diagnosis result corresponding to the sample feature vector.

[0112] In an embodiment of the present invention, the processing of the feature map to be used by the fully connected layer to obtain the fault diagnosis result corresponding to the sample feature vector includes: processing the feature map to be used by the fully connected layer to map it to the sample label space, and then calculating the predicted probability value by the softmax activation function; and using the predicted probability value and the predicted label corresponding to the predicted probability value as the fault diagnosis result.

[0113] The predicted label may be a fault diagnosis result predicted by the initial battery fault diagnosis model, and the predicted probability value may be a confidence level corresponding to the predicted label, etc.

[0114] Specifically, Figure 7 As shown, this is the overall structure diagram of the target battery fault diagnosis model provided in Example 2 of the present invention. After feature extraction by the three modules of multi-scale time convolution module, hybrid attention mechanism module and Ghost module, the learned distributed features are mapped to the sample label space using the fully connected layer, and then the probability value is calculated by the softmax activation function to obtain the fault diagnosis result.

[0115] S270: Correct the model parameters of the initial battery fault diagnosis model according to the fault diagnosis result and the fault label value to obtain the target battery fault diagnosis model.

[0116] It is understandable that the fault diagnosis results predicted by the initial battery fault diagnosis model may deviate from the true value, so the initial battery fault diagnosis model can be trained iteratively through a large number of samples. That is, each sample feature vector and the corresponding fault label can be used as a training sample to train the initial battery fault diagnosis model, and the target battery fault diagnosis model is obtained when the training is completed.

[0117] In a preferred embodiment, the feature extraction method of the present invention and the established target battery fault diagnosis model can be used to carry out hydrogen fuel cell fault diagnosis experiments. The experimental results show that the use of two modal decomposition algorithms can make up for the shortcomings of a single decomposition algorithm and extract the fault characteristics of the MEA voltage signal as much as possible. In addition, the established target battery fault diagnosis model can extract features from multiple scales, effectively utilizing the complementary capabilities of shallow and deep features, and can capture subtle fault characteristics from the PEMFC voltage signal, greatly improving the accuracy of PEMFC fault identification.

[0118] The loss and accuracy curves of the established target battery fault diagnosis model during training are shown in the figure below: Figure 8 and Fig. 9As shown, it can be seen that the established target battery fault diagnosis model can effectively learn fault feature information when facing PEMFC fault diagnosis. Its loss function shows a decreasing trend during the training process and there is no large fluctuation. In addition, during the training process, its recognition accuracy can increase rapidly with the increase in the number of iterations and can converge quickly.

[0119] The following are the classification results of the test set for five working conditions: no fault, flooding fault, dehydration fault, air starvation fault, and hydrogen starvation fault. The proposed method can effectively identify various types of faults, and the recognition accuracy of various fault conditions reaches 99.8%.

[0120]

[0121] The comparison results with other neural networks are as follows. It can be seen that the established target battery fault diagnosis model has the highest accuracy for PEMFC fault diagnosis compared with other comparison neural networks, and can effectively identify various types of faults.

[0122]

[0123] Only by changing the modal decomposition algorithm and using the same original data samples and neural network, the comparison results of different feature extraction methods are obtained as follows. It can be seen that after combining the two modal decomposition algorithms and using the established neural network for fault diagnosis, the fault identification accuracy can be greatly improved at a lower cost in algorithm calculation time.

[0124]

[0125]

[0126] Embodiment 3

[0127] Fig.10 A flowchart of a battery fault diagnosis method provided in Example 3 of the present invention. This embodiment can be applied to the situation of fault diagnosis of proton exchange membrane fuel cells. The method can be executed by a battery fault diagnosis device, which can be implemented in the form of hardware and / or software, and the device can be configured in a computer device.

[0128] As shown in 10, the method includes:

[0129] S310, obtaining a voltage signal of a membrane electrode assembly to be diagnosed of a battery to be diagnosed.

[0130] The battery to be diagnosed may be a proton exchange membrane fuel cell that needs to be diagnosed for fault, and the membrane electrode assembly voltage signal to be diagnosed refers to the membrane electrode assembly voltage signal of the battery.

[0131] S320, processing the voltage signal of the membrane electrode assembly to be diagnosed by variational mode decomposition and integrated empirical mode decomposition to obtain a characteristic vector to be diagnosed of the battery to be diagnosed.

[0132] The feature vector to be diagnosed may be a feature vector obtained after being processed by variational mode decomposition and integrated empirical mode decomposition.

[0133] S330, inputting the feature vector to be diagnosed into a target battery fault diagnosis model to obtain a target diagnosis result corresponding to the battery to be diagnosed output by the target battery fault diagnosis model.

[0134] Specifically, the feature vector to be diagnosed is input into the target battery fault diagnosis model, and the target battery fault diagnosis model can output the diagnosis result of the battery to be diagnosed, that is, the target diagnosis result. A fault diagnosis method based on modal decomposition combined with a neural network uses modal decomposition to mine feature information hidden in the data, and further uses a neural network model to implement PEMFC fault diagnosis on data features.

[0135] The technical solution of the embodiment of the present invention obtains the voltage signal of the membrane electrode assembly to be diagnosed of the battery to be diagnosed, processes the voltage signal of the membrane electrode assembly to be diagnosed by variational mode decomposition and integrated empirical mode decomposition, and obtains the feature vector to be diagnosed of the battery to be diagnosed. It can effectively mine the feature information hidden in the data, and solve the problem that the fault distinguishability of the time domain, frequency domain or shallow features in the membrane electrode assembly voltage signal is not prominent, resulting in the low fault diagnosis accuracy of the current related algorithms. The sample feature vector is further used to train and classify the neural network model to establish a target battery fault diagnosis model. The model can improve the fault diagnosis accuracy and effectively identify various fault types of hydrogen fuel cells.

[0136] Embodiment 4

[0137] Fig.11 A schematic diagram of the structure of a battery fault diagnosis model training device provided in Embodiment 4 of the present invention.

[0138] like Fig.11 As shown, the device comprises:

[0139] A voltage signal acquisition module 410 is used to acquire a voltage signal of a membrane electrode assembly of a proton exchange membrane fuel cell and a fault label value corresponding to the voltage signal of the membrane electrode assembly;

[0140] A voltage signal decomposition module 420 is used to perform variational mode decomposition and integrated empirical mode decomposition on the membrane electrode assembly voltage signal to obtain a sample feature vector corresponding to the membrane electrode assembly voltage signal;

[0141] The model training module 430 is used to train an initial battery fault diagnosis model based on the sample feature vector and the fault label value to obtain a target battery fault diagnosis model, wherein the initial battery fault diagnosis model includes a multi-scale time convolution module, a hybrid attention mechanism module and a Ghost module.

[0142] Optionally, the voltage signal acquisition module 410 is specifically used to:

[0143] By using a proton exchange membrane fuel cell test platform, the proton exchange membrane fuel cell is placed in various operating conditions;

[0144] Acquiring a to-be-processed voltage signal of the membrane electrode assembly of the proton exchange membrane fuel cell under various working conditions, and performing normalization processing on the to-be-processed voltage signal to obtain the membrane electrode assembly voltage signal;

[0145] The type of operating condition includes at least one of a normal operating condition, a flooding fault, a dehydration fault, an air starvation fault and a hydrogen starvation fault.

[0146] Optionally, the voltage signal decomposition module 420 is specifically used to:

[0147] Decomposing the membrane electrode assembly voltage signal into a plurality of natural mode components by variational mode decomposition;

[0148] Decomposing the membrane electrode assembly voltage signal into a plurality of intrinsic mode components by integrated empirical mode decomposition;

[0149] An energy value of each of the intrinsic modal components is determined, and the sample feature vector is constructed based on the energy value of each of the extrinsic modal components.

[0150] Optionally, the model training module 430 includes:

[0151] A multi-scale time convolution module, used for processing the sample feature vector and outputting a first feature vector;

[0152] A hybrid attention mechanism module, used for processing the first feature vector and outputting a second feature vector corresponding to the first feature vector; wherein the hybrid attention mechanism module includes spatial attention and squeeze and excitation attention;

[0153] A Ghost module, used for processing the second feature vector and outputting a feature map to be used;

[0154] A fully connected layer module is used to process the feature map to be used and obtain a fault diagnosis result corresponding to the sample feature vector;

[0155] The model parameter correction module is used to correct the model parameters of the initial battery fault diagnosis model according to the fault diagnosis result and the fault label value to obtain the target battery fault diagnosis model.

[0156] Optionally, the multi-scale temporal convolution module includes three 1×1 convolution layers, two 3x3 convolution layers and one 5x5 convolution layer;

[0157] The multi-scale time convolution module is specifically used for:

[0158] Reducing the dimension of the sample feature vector through two 1×1 convolutional layers respectively to obtain a first reduced dimension feature vector and a second reduced dimension feature vector;

[0159] Performing temporal feature information on the first dimension reduction feature vector through two 3x3 convolutional layers, one 5x5 convolutional layer and one 1×1 convolutional layer to obtain a temporal feature vector;

[0160] The time feature vector is concatenated with the second dimension-reduced feature vector to obtain the first feature vector.

[0161] Optionally, the hybrid attention mechanism module is specifically used for:

[0162] Inputting the first feature vector into spatial attention to obtain a spatial attention feature vector;

[0163] Inputting the spatial attention feature vector into the squeeze and stimulate attention to obtain a squeeze and stimulate attention feature vector;

[0164] The spatial attention feature vector and the squeeze and incentive attention feature vector slice are concatenated to obtain the second feature vector.

[0165] Optionally, the fully connected layer module is specifically used for:

[0166] The feature map to be used is processed through the fully connected layer to map it to the sample label space, and then the predicted probability value is calculated through the softmax activation function;

[0167] The predicted probability value and the predicted label corresponding to the predicted probability value are used as the fault diagnosis result.

[0168] The battery fault diagnosis model training device provided in the embodiment of the present invention can execute the battery fault diagnosis model training method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0169] Embodiment 5

[0170] Fig.12A schematic diagram of the structure of an electronic device provided for Embodiment 5 of the present invention. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0171] like Fig.12 As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0172] A number of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0173] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The processor 11 executes the various methods and processes described above, such as a battery fault diagnosis model training method or a battery fault diagnosis method.

[0174] In some embodiments, the battery fault diagnosis model training method or the battery fault diagnosis method may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the battery fault diagnosis model training method or the battery fault diagnosis method described above may be performed. Alternatively, in other embodiments, the processor 11 may be configured to execute the battery fault diagnosis model training method or the battery fault diagnosis method in any other appropriate manner (e.g., by means of firmware).

[0175] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0176] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0177] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, 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), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0178] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0179] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0180] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0181] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0182] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A battery fault diagnosis model training method, characterized in that: include: Acquire a membrane electrode assembly voltage signal of a proton exchange membrane fuel cell and a fault label value corresponding to the membrane electrode assembly voltage signal; Performing variational mode decomposition and integrated empirical mode decomposition on the membrane electrode assembly voltage signal to obtain a sample feature vector corresponding to the membrane electrode assembly voltage signal; An initial battery fault diagnosis model is trained based on the sample feature vector and the fault label value to obtain a target battery fault diagnosis model, wherein the initial battery fault diagnosis model includes a multi-scale time convolution module, a hybrid attention mechanism module and a Ghost module.

2. The method according to claim 1, characterized in that The method of obtaining a voltage signal of a membrane electrode assembly of a proton exchange membrane fuel cell comprises: By using a proton exchange membrane fuel cell test platform, the proton exchange membrane fuel cell is placed in various operating conditions; Acquiring a to-be-processed voltage signal of the membrane electrode assembly of the proton exchange membrane fuel cell under various working conditions, and performing normalization processing on the to-be-processed voltage signal to obtain the membrane electrode assembly voltage signal; The type of operating condition includes at least one of a normal operating condition, a flooding fault, a dehydration fault, an air starvation fault and a hydrogen starvation fault.

3. The method according to claim 1, characterized in that The performing variational mode decomposition and integrated empirical mode decomposition on the membrane electrode assembly voltage signal to obtain a sample feature vector corresponding to the membrane electrode assembly voltage signal includes: Decomposing the membrane electrode assembly voltage signal into a plurality of natural mode components by variational mode decomposition; Decomposing the membrane electrode assembly voltage signal into a plurality of intrinsic mode components by integrated empirical mode decomposition; An energy value of each of the intrinsic modal components is determined, and the sample feature vector is constructed based on the energy value of each of the extrinsic modal components.

4. The method according to claim 1, characterized in that: The initial battery fault diagnosis model is trained based on the sample feature vector and the fault label value to obtain a target battery fault diagnosis model, including: Inputting the sample feature vector into the multi-scale time convolution module to obtain a first feature vector output by the multi-scale time convolution module; Inputting the first feature vector into the hybrid attention mechanism module to obtain a second feature vector corresponding to the first feature vector; wherein the hybrid attention mechanism module includes spatial attention and squeeze and incentive attention; Inputting the second feature vector into the Ghost module to obtain a feature map to be used output by the Ghost module; Processing the feature map to be used through a fully connected layer to obtain a fault diagnosis result corresponding to the sample feature vector; The model parameters of the initial battery fault diagnosis model are modified according to the fault diagnosis result and the fault label value to obtain the target battery fault diagnosis model.

5. The method according to claim 4, characterized in that The multi-scale temporal convolution module includes three 1×1 convolutional layers, two 3x3 convolutional layers and one 5x5 convolutional layer; The step of inputting the sample feature vector into the multi-scale time convolution module to obtain a first feature vector output by the multi-scale time convolution module includes: Reducing the dimension of the sample feature vector through two 1×1 convolutional layers respectively to obtain a first reduced dimension feature vector and a second reduced dimension feature vector; Performing temporal feature information on the first dimension reduction feature vector through two 3x3 convolutional layers, one 5x5 convolutional layer and one 1×1 convolutional layer to obtain a temporal feature vector; The time feature vector is concatenated with the second dimension-reduced feature vector to obtain the first feature vector.

6. The method according to claim 5, characterized in that The step of inputting the first feature vector into the hybrid attention mechanism module to obtain a second feature vector corresponding to the first feature vector includes: Inputting the first feature vector into spatial attention to obtain a spatial attention feature vector; Inputting the spatial attention feature vector into the squeeze and stimulate attention to obtain a squeeze and stimulate attention feature vector; The spatial attention feature vector and the squeeze and incentive attention feature vector slice are concatenated to obtain the second feature vector.

7. The method according to claim 6, characterized in that The processing of the feature map to be used by the fully connected layer to obtain the fault diagnosis result corresponding to the sample feature vector includes: The feature map to be used is processed through the fully connected layer to map it to the sample label space, and then the predicted probability value is calculated through the softmax activation function; The predicted probability value and the predicted label corresponding to the predicted probability value are used as the fault diagnosis result.

8. A battery fault diagnosis method, characterized in that: The method comprises: Acquiring a voltage signal of a membrane electrode assembly to be diagnosed of a battery to be diagnosed; Processing the voltage signal of the membrane electrode assembly to be diagnosed by variational mode decomposition and integrated empirical mode decomposition to obtain a characteristic vector to be diagnosed of the battery to be diagnosed; The feature vector to be diagnosed is input into the target battery fault diagnosis model described in any one of claims 1 to 7 to obtain a target diagnosis result corresponding to the battery to be diagnosed output by the target battery fault diagnosis model.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the battery fault diagnosis model training method described in any one of claims 1-7, or the battery fault diagnosis method described in claim 8.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the battery fault diagnosis model training method described in any one of claims 1 to 7, or the battery fault diagnosis method described in claim 8 when executed.

Images