Fuel cell hydrogen leakage fault diagnosis method based on VAE model and SVM model

Abstract

The invention relates to the technical field of fuel cell fault diagnosis, in particular to a fuel cell hydrogen leakage fault diagnosis method based on a VAE model and an SVM model, and the method comprises the following steps: collecting the original data of the operation of a fuel cell, and dividing the original data into a normal data set and a fault data set; preprocessing the normal data set and the fault data set to obtain a normal data sample and a fault data sample; constructing and training a VAE model, and expanding the trained VAE model to obtain a synthesized fault data sample; combining the data samples to form an expanded data set; performing normalization processing, extracting frequency domain features and time domain features of each data sample, and performing singular value decomposition processing to obtain a key feature vector set; constructing and training an SVM model, optimizing parameters of the SVM model by using an improved BKA algorithm, and performing fault diagnosis by using the trained SVM model; according to the invention, imbalance between hydrogen leakage data and normal data is eliminated, and hydrogen leakage is rapidly and accurately identified.

Description

Technical Field

[0001] The present invention relates to the technical field of fuel cell fault diagnosis, and particularly to a fuel cell hydrogen leakage fault diagnosis method based on a VAE model and an SVM model. Background Technique

[0002] As an efficient clean energy conversion technology, the monitoring and fault diagnosis of the operating state of a fuel cell system are of great significance for ensuring system stability and extending service life. The traditional method for diagnosing the hydrogen leakage fault of a fuel cell system mainly relies on hydrogen sensors to collect hydrogen concentration data, and then analyzes it in combination with manual experience. This diagnosis method not only hardly meets the development needs of intelligence and automation, but also limits the accuracy of fault diagnosis due to over-reliance on manual judgment and single data, while increasing labor and time costs.

[0003] At present, machine learning algorithms are also used for the fault diagnosis of fuel cells. Since machine learning algorithms rely on a large amount of data training, and the occurrence of faults in the actual operation of a fuel cell system is usually a small probability event, coupled with imperfect data records, the number of fault samples is extremely limited, and the ratio of normal operation data to fault data is seriously unbalanced, resulting in insufficient diagnostic ability of machine learning algorithms for fault samples. Especially in the scenario of multiple fault types, high misdiagnosis rates or missed diagnosis rates are very likely to occur, and fault identification is not timely enough. In a fuel cell system, the missed judgment of a fault signal may directly lead to a decline in system performance or even failure, and its consequences are far more serious than the misjudgment of a normal device. Especially in key fault scenarios such as hydrogen leakage, missed judgment may not only lead to serious safety accidents, but also cause system damage.

[0004] Therefore, in the case of unbalanced data distribution in a fuel cell system, it is very necessary to develop a diagnostic method that can quickly and accurately identify hydrogen leakage. Summary of the Invention

[0005] The technical problem to be solved by the present invention is: to overcome the deficiencies in the prior art and provide a fuel cell hydrogen leakage fault diagnosis method based on a VAE model and an SVM model that eliminates the imbalance between hydrogen leakage data and normal data and can quickly and accurately identify hydrogen leakage.

[0006] The technical solution adopted by the present invention to solve its technical problems is: a fuel cell hydrogen leakage fault diagnosis method based on a VAE model and an SVM model, including the following steps:

[0007] S1. Collect the original data of the fuel cell operation, and divide the original data into a normal data set and a fault data set;

[0008] S2. Preprocess the normal dataset and the fault dataset to obtain normal data samples and fault data samples;

[0009] S3. Construct a VAE model, train the VAE model using the fault data samples, and use the trained VAE model for data augmentation to obtain synthetic fault data samples;

[0010] S4. Combine the synthetic fault data samples, the fault data samples, and the normal data samples to form an augmented dataset, and divide the augmented dataset into a training set and a test set;

[0011] S5. Normalize the training set and the test set, extract the frequency-domain features and time-domain features of each data sample in the training set and the test set, and then perform singular value decomposition on the frequency-domain features and the time-domain features to obtain the key feature vector sets of the training set and the test set;

[0012] S6. Construct an SVM model, train the SVM model with the key feature vector set of the training set, and at the same time optimize the parameters of the SVM model using an improved BKA algorithm, and use the trained SVM model for fault diagnosis.

[0013] Furthermore, the VAE model in step S3 includes an encoder, a decoder, and an optimizer. The encoder maps the input data to the latent space, the decoder maps the low-dimensional representation in the latent space back to the data space, and the optimizer optimizes the VAE model according to the reconstruction loss and KL divergence in the encoder and the decoder.

[0014] Furthermore, the training process of the VAE model in step S3 is as follows:

[0015] S31. The encoder receives the fault data samples and calculates the mean u φ (z / x) and variance φ (x) of the conditional distribution Q The conditional distribution Q φ (z / x) is as follows:

[0016]

[0017] where z is the low-dimensional representation of the input data mapped to the latent space by the encoder; x is the input data of the encoder; φ is the parameter;

[0018] S32. Introduce the reparameterization trick, sample random noise ε from the standard normal distribution N(0, I), and combine the mean u φ (x) and variance to obtain the low-dimensional representation z of the latent space, and the formula is as follows:

[0019]

[0020] In the formula, ⊙ represents element-wise multiplication;

[0021] S33. The decoder maps the low-dimensional representation z of the latent space back to the data space to generate the reconstructed data probability distribution P θ (x / z), and the probability distribution P θ (x / z) is as follows:

[0022]

[0023] In the formula, is the output of the decoder; σ is the standard deviation of the reconstructed data distribution; I is the covariance matrix of the reconstructed data .

[0024] Furthermore, the loss function L VAE in the VAE model is as follows:

[0025] L VAE = E Qφ(z / x) [logP θ (x / z)] - D KL [Q φ (z / x) || p(z)]

[0026] In the formula, E Qφ(z / x) [] is the reconstruction loss, which is used to measure the difference between the reconstructed data and the input data x; D KL [] is the KL divergence, which is used to measure the difference between the conditional distribution Q φ (z / x) and the prior distribution p(z), and p(z) is the standard normal distribution N(0, I).

[0027] Furthermore, the calculation process of the SVM model in step S6 is as follows:

[0028] Set the hyperplane function f(x) = ω T x + b, and solve for the optimal classification plane. The solution formula is as follows:

[0029]

[0030] s.t. y i (ω T x i + b) ≥ 1

[0031] In the formula, ω is the weight vector; x i is the key feature vector; y i is the label of different state types; b is the bias vector; i = 1, 2,..., n;

[0032] Introduce a penalty coefficient $c$ to control the tolerance of classification errors, and the solution formula is transformed into the following formula:

[0033]

[0034] s.t. $y$ i $(ω$ T $x$ i $+b)≥1 - Δ$ i

[0035] where $c≥0$; $Δ$ i is a slack variable used to tolerate abnormal data, and $Δ$ i $≥0$;

[0036] Use the Lagrangian function for solution. According to the KKT conditions, transform the problem of solving the optimal plane into a quadratic optimization problem. The formula for the quadratic optimization problem is as follows:

[0037]

[0038] where $α$ i and $α$ j are Lagrange multipliers, $0≤α$ i $≤c$, $0≤α$ j $≤c$;

[0039] The optimal classification plane function obtained by solving the quadratic optimization problem is:

[0040]

[0041] where $sgn{}$ is the sign function, $α$ i · is the optimal Lagrange multiplier, and $b$ · is the optimal bias vector;

[0042] Introduce the Gaussian kernel function $k(x$ i , $x$ j ) to map low-dimensional data to a high-dimensional space to achieve linear separability. The optimal classification plane function is transformed into:

[0043]

[0044] where $k(x$ i , $x$ j ) = $exp(-g||x$ i $-x$ j $||$ 2 ),

[0045] where $g$ is the parameter of the Gaussian kernel function, which determines the number of support vectors in the high-dimensional space and affects the training speed; $ρ$ is the bandwidth parameter of the Gaussian kernel function; $||x$ i-x j || 2 is x i and x j the distance between.

[0046] Furthermore, the steps of optimizing the SVM model parameters by the improved BKA algorithm in step S6 are as follows:

[0047] S61. Initialize the parameters of the BKA algorithm, set the search ranges of the penalty coefficient c and the Gaussian kernel function parameter g, and encode them as the individual positions, and randomly generate the initial population;

[0048] S62. Use the test set to evaluate the fault diagnosis accuracy of each individual corresponding SVM model as the fitness value, and set the individual position with the highest fitness value in the initial population as the global optimal position;

[0049] S63. The individual positions are iteratively adjusted according to the update formula, and the update formula is as follows:

[0050]

[0051] where,

[0052] In the formula, L(λ) is the sampling step size in the Lévy distribution; λ is the exponential parameter of the Lévy distribution, λ ∈ (1, 2); W1 and W2 are the parent individuals; cd1 and cd2 are the offspring individuals; β is a random vector, β ∈ [0, 1]; u and v are random numbers sampled from the standard normal distribution;

[0053] S64. Record the individual optimal position and the global optimal position in each iteration to guide the optimization direction. If the iteration meets the termination condition, output the optimal parameter combination of the penalty coefficient c and the Gaussian kernel function parameter g.

[0054] Furthermore, the differential evolution algorithm is used for preprocessing the normal data set and the fault data set in step S2.

[0055] Furthermore, in step S5, Fourier transform is performed on each data sample in the training set and the test set to extract frequency domain features, and at the same time, basic statistics are used to extract time domain features.

[0056] Furthermore, the expanded data set in step S4 is divided into a 70% training set and a 30% test set.

[0057] The beneficial effects of the present invention are:

[0058] (1) The present invention expands the fault data samples through the VAE model, effectively solving the problem of data imbalance caused by a large amount of normal data samples and a small amount of hydrogen leakage fault data samples in the fuel cell operation data, ensuring that the SVM model can be fully and effectively trained, and significantly improving the accuracy of the SVM model in identifying hydrogen leakage faults; combined with the use of an improved BKA algorithm to optimize the parameters of the SVM model (i.e., the penalty coefficient c and the Gaussian kernel function parameter g), further improving the accuracy of the SVM model in identifying hydrogen leakage faults and greatly reducing the diagnosis time.

[0059] (2) The present invention extracts frequency-domain features and time-domain features to remove some redundant and irrelevant features in the training set and the test set, enhancing the richness of the input information of the SVM model and improving the classification accuracy and generalization performance of the SVM model. Then, singular value decomposition is used to process the extracted frequency-domain features and time-domain features, reducing the number of features and eliminating the interference of redundant features. Finally, hydrogen leakage fault diagnosis is carried out through key features, improving the diagnosis efficiency and further reducing the diagnosis time.

[0060] (3) Levy flight is introduced into the improved BKA algorithm of the present invention. Due to the long-distance jumping characteristic of Levy flight, it can significantly enhance the global search ability of the algorithm, thus helping the algorithm effectively jump out of the local optimal solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] The present invention will be further described below in conjunction with the drawings and embodiments.

[0062] Figure 1 is the flowchart of the steps of the present invention;

[0063] Figure 2 is the diagnostic effect diagram of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0064] The present invention will now be further described in conjunction with the drawings and preferred embodiments. These drawings are all simplified schematic diagrams, only illustrating the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0065] As Figure 1 shown, a fuel cell hydrogen leakage fault diagnosis method based on the VAE model and the SVM model includes the following steps:

[0066] S1. Collect the original data of the fuel cell operation and divide the original data into a normal data set and a fault data set. Specifically, the original data includes operating parameters such as the current, voltage, temperature of the stack, and the hydrogen concentration in the environment.

[0067] S2. Preprocess the normal dataset and the fault dataset to obtain normal data samples and fault data samples. Specifically, the differential evolution algorithm is used for preprocessing the normal dataset and the fault dataset.

[0068] S3. Construct a VAE model, train the VAE model using the fault data samples, and perform data augmentation using the trained VAE model to obtain synthetic fault data samples.

[0069] S4. Combine the synthetic fault data samples, the fault data samples, and the normal data samples to form an augmented dataset, and divide the augmented dataset into a training set and a test set. Specifically, the augmented dataset is divided into a 70% training set and a 30% test set.

[0070] S5. Perform normalization processing on the training set and the test set, extract the frequency-domain features and time-domain features of each data sample in the training set and the test set, and then perform singular value decomposition processing on the frequency-domain features and time-domain features to obtain the key feature vector sets of the training set and the test set. Specifically, perform Fourier transform on each data sample in the training set and the test set to extract frequency-domain features; use basic statistics to extract time-domain features.

[0071] Through the extraction of frequency-domain features and time-domain features, some redundant and irrelevant features in the training set and the test set are removed, enhancing the richness of the input information of the SVM model and improving the classification accuracy and generalization performance of the SVM model. Then, perform singular value decomposition processing on the extracted frequency-domain features and time-domain features to reduce the number of features and eliminate the interference of redundant features. Finally, perform hydrogen leakage fault diagnosis through the key features, improving the diagnosis efficiency and further reducing the diagnosis time.

[0072] S6. Construct an SVM model, train the SVM model using the key feature vector set of the training set, and at the same time optimize the parameters of the SVM model using the improved BKA algorithm, and perform fault diagnosis using the trained SVM model.

[0073] By augmenting the fault data samples through the VAE model (variational autoencoder model), the problem of data imbalance caused by a large amount of normal data samples and a small amount of hydrogen leakage fault data samples in the fuel cell operation data is effectively solved, ensuring that the SVM model (support vector machine model) can be fully and effectively trained, and significantly improving the accuracy of the SVM model in identifying hydrogen leakage faults; combined with the use of the improved BKA algorithm (blackcap warbler algorithm) to optimize the parameters of the SVM model (i.e., the penalty coefficient c and the Gaussian kernel function parameter g), the accuracy of the SVM model in identifying hydrogen leakage faults is further improved, and the diagnosis time is greatly reduced.

[0074] In step S3, the VAE model includes an encoder, a decoder, and an optimizer. The encoder maps the input data to the latent space, the decoder maps the low-dimensional representation in the latent space back to the data space, and the optimizer optimizes the VAE model according to the reconstruction loss and KL divergence in the encoder and decoder.

[0075] The training process of the VAE model in step S3 is as follows:

[0076] S31. The encoder receives the fault data samples and calculates the mean μ φ (z / x) and variance φ σ² of the conditional distribution Q φ (z / x) as follows:

[0077]

[0078] where z is the low-dimensional representation of the input data mapped by the encoder to the latent space; x is the input data of the encoder; φ is the parameter;

[0079] S32. Introduce the reparameterization trick, sample the random noise ε from the standard normal distribution N(0, I), and combine the mean μ φ (x) and variance σ²

[0080]

[0081] to obtain the low-dimensional representation z of the latent space, and the formula is as follows:

[0082] S33. The decoder maps the low-dimensional representation z of the latent space back to the data space to generate the probability distribution P (x / z) of the reconstructed data θ (x / z), and the probability distribution P θ (x / z) is as follows:

[0083]

[0084] where is the output of the decoder; σ is the standard deviation of the probability distribution of the reconstructed data ; I is the covariance matrix of the reconstructed data . Specifically, I is the identity matrix.

[0085] The training of the VAE model is achieved by maximizing the evidence lower bound. By maximizing the evidence lower bound, the VAE model can learn the complex latent structure of the data and generate new data samples.

[0086] The loss function L VAE in the VAE model is as follows:

[0087] L VAE = E Qφ(z / x) [logP θ (x / z)] - D KL [Q φ (z / x) || p(z)]

[0088] In the formula, is the reconstruction loss, used to measure the difference between the reconstructed data and the input data x; D KL [] is the KL divergence, used to measure the difference between the conditional distribution Q φ (z / x) and the prior distribution p(z), where p(z) is the standard normal distribution N(0, I).

[0089] The calculation process of the SVM model in step S6 is as follows:

[0090] Set the hyperplane function f(x) = ω T x + b, and solve for the optimal classification plane. The solution formula is as follows:

[0091]

[0092] s.t. y i (ω T x i + b) ≥ 1

[0093] In the formula, ω is the weight vector; x i is the key feature vector; y i is the label of different state types; b is the bias vector; i = 1, 2,..., n.

[0094] By solving the optimal classification plane, the correct division of the training data is achieved and the data interval is maximized.

[0095] Introduce the penalty coefficient c to control the tolerance of classification errors, and the solution formula is transformed into the following formula:

[0096]

[0097] s.t. y i (ω T x i + b) ≥ 1 - Δ i

[0098] In the formula, c ≥ 0; Δ i is the slack variable, used to tolerate abnormal data, and Δ i ≥ 0.

[0099] Since the fuel cell system has non - linear dynamic characteristics and complex parameter variations, it is difficult to describe with simple linear relationships. Therefore, to cope with this complexity, this application introduces a penalty coefficient c into the SVM model to control the tolerance of classification errors and balance the complexity and accuracy of the model.

[0100] Use the Lagrangian function for solution. According to the KKT conditions (Karush - Kuhn - Tucker conditions), transform the problem of solving the optimal plane into a quadratic optimization problem. The formula for the quadratic optimization problem is as follows:

[0101]

[0102]

[0103] In the formula, α i and α j are Lagrange multipliers, 0 ≤ α i ≤ c, 0 ≤ α j ≤ c;

[0104] The optimal classification plane function obtained by solving the quadratic optimization problem is:

[0105]

[0106] In the formula, sgn{} is the sign function, is the optimal Lagrange multiplier, and b · is the optimal bias vector.

[0107] Introduce the Gaussian kernel function k(x i , x j ) to map the low - dimensional data to a high - dimensional space to achieve linear separability. The optimal classification plane function is transformed into:

[0108]

[0109] Among them, k(x i , x j ) = exp(-g||x i -x j || 2 ),

[0110] In the formula, g is the parameter of the Gaussian kernel function, which determines the number of support vectors in the high - dimensional space and affects the training speed; ρ is the bandwidth parameter of the Gaussian kernel function; ||x i -x j || 2 is the distance between x i and x j .

[0111] The steps of optimizing the SVM model parameters by the improved BKA algorithm in step S6 are as follows:

[0112] S61. Initialize the parameters of the BKA algorithm, set the search ranges of the penalty coefficient c and the Gaussian kernel function parameter g, encode them as individual positions, and randomly generate an initial population.

[0113] S62. Use the test set to evaluate the fault diagnosis accuracy of each individual corresponding to the SVM model as the fitness value, and set the individual position with the highest fitness value in the initial population as the global optimal position.

[0114] S63. The individual positions are iteratively adjusted according to the update formula, and the update formula is as follows:

[0115]

[0116] Among them,

[0117] In the formula, L(λ) is the sampling step size in the Lévy distribution; λ is the exponential parameter of the Lévy distribution, λ ∈ (1, 2); W1 and W2 are parent individuals; cd1 and cd2 are offspring individuals; β is a random vector, β ∈ [0, 1]; u and v are random numbers sampled from the standard normal distribution.

[0118] By introducing the sampling step size of the Lévy distribution, the positions of the offspring individuals are no longer limited to the linear combination range of the parent individuals, thereby increasing the diversity and exploration ability of the population.

[0119] S64. Record the individual optimal position and the global optimal position in each iteration to guide the optimization direction. If the iteration meets the termination condition, output the optimal parameter combination of the penalty coefficient c and the Gaussian kernel function parameter g.

[0120] In the improved BKA algorithm, Lévy flight is introduced. Due to the characteristic of long-distance jumps of Lévy flight, it can significantly enhance the global search ability of the algorithm, thereby helping the algorithm effectively jump out of the local optimal solution.

[0121] From Figure 2 it can be observed that the labels predicted by the fault diagnosis method proposed in this application are basically the same as the true labels, where 1 indicates correct diagnosis and 0 indicates incorrect diagnosis.

[0122] The above embodiments are only for explaining the technical concept and characteristics of the present invention, and their purpose is to enable those who are familiar with this technology to understand the content of the present invention and implement it. It cannot be used to limit the protection scope of the present invention. Any equivalent changes or modifications made according to the spirit and essence of the present invention should be covered within the protection scope of the present invention.

Claims

1. A fuel cell hydrogen leakage fault diagnosis method based on the VAE model and the SVM model, characterized in that, It includes the following steps: S1. Collect the original data of the fuel cell operation, and divide the original data into a normal data set and a fault data set; S2. Preprocess the normal data set and the fault data set to obtain normal data samples and fault data samples; S3. Construct a VAE model, train the VAE model using the fault data samples, and use the trained VAE model for data augmentation to obtain synthetic fault data samples; S4. Merge the synthetic fault data samples, the fault data samples and the normal data samples to form an augmented data set, and divide the augmented data set into a training set and a test set; S5. Perform normalization processing on the training set and the test set, extract the frequency domain features and time domain features of each data sample in the training set and the test set, and then perform singular value decomposition processing on the frequency domain features and the time domain features to obtain the key feature vector sets of the training set and the test set; S6. Construct an SVM model, train the SVM model with the key feature vector set of the training set, and at the same time optimize the parameters of the SVM model using an improved BKA algorithm, and use the trained SVM model for fault diagnosis.

2. The fuel cell hydrogen leakage fault diagnosis method based on the VAE model and the SVM model according to claim 1, wherein In step S3, the VAE model includes an encoder, a decoder and an optimizer. The encoder maps the input data to the latent space, the decoder maps the low-dimensional representation in the latent space back to the data space, and the optimizer optimizes the VAE model according to the reconstruction loss and KL divergence in the encoder and the decoder.

3. The fuel cell hydrogen leakage fault diagnosis method based on the VAE model and the SVM model according to claim 2, wherein, The training process of the VAE model in step S3 is as follows: S31. The encoder receives the fault data samples and calculates the conditional distribution Q of the latent space φ The mean u of φ (z / x) and the variance The conditional distribution Q φ (z / x) is as follows: In the formula, z is the low-dimensional representation that the encoder maps the input data to the latent space; x is the input data of the encoder; φ is the parameter; S32. Introduce the reparameterization trick, sample the random noise ε from the standard normal distribution N(0, I), and combine it with the mean μ φ (x) and the variance to obtain the low-dimensional representation z of the latent space, and the formula is as follows: In the formula, ⊙ is element-wise multiplication; S33. The decoder maps the low-dimensional representation z of the latent space back to the data space to generate the reconstructed data of the probability distribution P θ (x / z), the probability distribution P θ (x / z) is as follows: In the formula, is the output of the decoder; σ is the standard deviation of the probability distribution of the reconstructed data ; I is the covariance matrix of the reconstructed data .

4. The fuel cell hydrogen leakage fault diagnosis method based on the VAE model and the SVM model according to claim 3, characterized in that, The loss function L in the VAE model VAE is as follows: where is the reconstruction loss, used to measure the difference between the reconstructed data and the input data x; D KL [] is the KL divergence, used to measure the difference between the conditional distribution Q φ (z / x) and the prior distribution p(z), where p(z) is the standard normal distribution N(0, I).

5. The fuel cell hydrogen leakage fault diagnosis method based on the VAE model and the SVM model according to claim 1, characterized in that The calculation process of the SVM model in step S6 is as follows: Set the hyperplane function f(x) = ω T x + b, solve for the optimal classification plane, and the solution formula is as follows: s.t.y i (ω T x i +b)≥1 where ω is the weight vector; x i is the key feature vector; y i is the label of different state types; b is the bias vector; i = 1, 2,..., n; Introduce a penalty coefficient c to control the tolerance of classification errors, and the solution formula is transformed into the following formula: s.t.y i (ω T x i +b)≥1-Δ i where c ≥ 0; Δ i is a slack variable used to tolerate abnormal data, and Δ i ≥ 0; Use the Lagrangian function for solution, and according to the KKT conditions, transform the problem of solving the optimal plane into the problem of solving a quadratic optimization problem. The quadratic optimization problem formula is as follows: where α i and α j are Lagrange multipliers, 0 ≤ α i ≤ c, 0 ≤ α j ≤ c; The optimal classification plane function obtained by solving the quadratic optimization problem is: where sgn{} is the sign function, is the optimal Lagrange multiplier, and b · is the optimal bias vector; Introduce the Gaussian kernel function k(x i , x j ) to map the low-dimensional data to a high-dimensional space to achieve linear separability, and the optimal classification plane function is transformed into: where k(x i , x j ) = exp(-g||x i - x j || 2 ), where \(g\) is the parameter of the Gaussian kernel function, which determines the number of support vectors in the high-dimensional space and affects the training speed; \(\rho\) is the bandwidth parameter of the Gaussian kernel function; \(\left\|\mathbf{x} i -\mathbf{x} j \right\| 2 is the distance between \(\mathbf{x} i and \(\mathbf{x} j .

6. The fuel cell hydrogen leakage fault diagnosis method based on the VAE model and the SVM model according to claim 5, wherein The steps of optimizing the parameters of the SVM model by the improved BKA algorithm in step S6 are as follows: S61. Initialize the parameters of the BKA algorithm, set the search ranges of the penalty coefficient c and the Gaussian kernel function parameter g, and encode them as individual positions, and randomly generate an initial population; S62. Use the test set to evaluate the fault diagnosis accuracy of each individual corresponding SVM model as the fitness value, and set the individual position with the highest fitness value in the initial population as the global optimal position; S63. The individual positions are iteratively adjusted according to the update formula. The update formula is as follows: Among them, In the formula, L(λ) is the sampling step size in the Lévy distribution; λ is the exponential parameter of the Lévy distribution, λ ∈ (1, 2); W1 and W2 are parent individuals; cd1 and cd2 are offspring individuals; β is a random vector, β ∈ [0, 1]; u and v are random numbers sampled from the standard normal distribution; S64. Record the individual optimal position and the global optimal position in each iteration to guide the optimization direction. If the iteration meets the termination condition, output the optimal parameter combination of the penalty coefficient c and the Gaussian kernel function parameter g.

7. The fuel cell hydrogen leakage fault diagnosis method based on the VAE model and the SVM model according to claim 1, characterized in that In the step S2, the differential evolution algorithm is used for preprocessing the normal data set and the fault data set.

8. The fuel cell hydrogen leakage fault diagnosis method based on the VAE model and the SVM model according to claim 1, wherein In the step S5, Fourier transform is performed on each data sample in the training set and the test set to extract frequency domain features, and at the same time, basic statistics are used to extract time domain features.

9. The fuel cell hydrogen leakage fault diagnosis method based on the VAE model and the SVM model according to claim 1, wherein In the step S4, the augmented data set is divided into a training set of 70% and a test set of 30%.

Images