Battery swap station equipment fault diagnosis method and device, battery swap station, and storage medium

Abstract

The application relates to a battery swap station equipment fault diagnosis method, which comprises the following steps: performing concept representation processing on equipment operation data to obtain a concept representation matrix; performing concept representation processing on a fault sample generated according to a fault record to obtain a fault representation matrix about the fault record; and determining the distance between the concept representation matrix and the fault representation matrix, and determining whether a fault exists according to the distance.

Description

TECHNICAL FIELD

[0001] The present application relates to fault diagnosis of battery swap stations, in particular, to a battery swap station equipment fault diagnosis method and device, a battery swap station, and a storage medium. BACKGROUND

[0002] A battery swap station is an important energy supplement infrastructure for new energy vehicles. Faults occurring during the battery swap process of the battery swap station equipment will affect the battery swap experience and may cause safety risks to the vehicle, the battery, and other equipment. There are many types of equipment in a battery swap station, and there are many fault modes corresponding to different types of equipment. At present, the method for diagnosing faults of the battery swap station equipment still mainly relies on human experience, and fault features can only be extracted from single-variable time series signals. Due to the particularity of the battery swap station equipment, multiple signals (such as torque signals, speed signals, and displacement signals) are generally used to detect the operation of the equipment at the same time. Therefore, for the battery swap station equipment, the current method cannot extract features of complex fault modes with long-term, non-steady dynamic characteristics of the battery swap station equipment, and thus cannot perform accurate fault diagnosis.

[0003] With the development of the field of fault diagnosis, different kinds of signal processing methods have been used for fault feature extraction from signals, for example, Empirical Mode Decomposition (EMD) (see Lei YG, Lin J, He ZJ, Zuo MJ. A review on empirical mode decomposition in fault diagnosis of rotating machinery. Mech Syst Signal Process 2013; 35: 108-126), Wavelet Transform (WT) (see Lin J, Qu LS. Feature extraction based on Morlet wavelet and its application for mechanical fault diagnosis. J Sound Vib 2000; 234: 135-148 and Peng ZK, Chu FL. Application of the wavelet transform in machine condition monitoring and fault diagnostics: a review with bibliography. Mech Syst Signal Process 2004; 18: 199-221), Spectral Kurtosis (SC) (see Antoni J, Randall RB. The spectral kurtosis: application to the vibratory surveillance and diagnostics of rotating machines. Mech Syst Signal Process 2006; 20: 308-331) and others have been used for fault feature extraction from signals.

[0004] On the other hand, neural network technology is also applied to feature extraction of signals. For example, sparse coding (see Jiao, J., Zhao, M., Lin, J., & Liang, K. (2019). Hierarchical discriminating sparse coding for weak fault feature extraction of rolling bearings. Reliability Engineering & System Safety, 184, 41-54), autoencoder (see R. Thirukovalluru, S. Dixit, R. K. Sevakula, N. K. Verma, and A. Salour, “Generating feature sets for fault diagnosis using denoising stacked autoencoder,” in Proc. IEEE Int. Conf. Prognostics Health Manage. (ICPHM), Jun. 2016, pp. 1-7), variational autoencoder, etc. are used to extract feature signals.

[0005] Although the signal processing and neural network methods can perform feature extraction on device time series, the signal processing method cannot perform multivariate time series feature extraction, and the neural network method relies on a large number of fault samples for training. For the unique scenario of the battery swap station, that is, monitoring multiple signals of the device such as torque signals, displacement signals and speed signals, but the number of fault samples is very small or even only a single fault sample.

[0006] The signal processing and neural network methods cannot simultaneously meet these characteristics, and thus cannot extract effective fault features, which in turn will generate a large number of false alarms. In view of this, an improved fault detection mechanism needs to be proposed. SUMMARY

[0007] Embodiments of the present application provide a battery swap station device fault diagnosis method and device, a battery swap station and a storage medium, for detecting whether a battery swap station device has a fault.

[0008] According to one aspect of this application, a method for diagnosing equipment faults in a battery swapping station is provided. The method includes the following steps: performing conceptual characterization processing on equipment operating data to obtain a conceptual characterization matrix; performing conceptual characterization processing on fault samples generated based on fault records to obtain a fault characterization matrix for the fault records; and determining the distance between the conceptual characterization matrix and the fault characterization matrix, and determining whether a fault exists based on the distance.

[0009] Optionally, in some embodiments of this application, the method further includes: preprocessing the time-series data generated by the device operation to generate the device operation data.

[0010] In some embodiments of this application, optionally, the time series data u = [u1, u2, ..., u...] n ], and u i This represents the i-th time series sample in the time series data u, where 1 ≤ i ≤ n and i ∈ N+. The preprocessing includes standardization or normalization of the time series data, wherein: the standardization includes global standardization and sample standardization, where: u i The j-th element u i，j The global standardized value is u i，j The ratio of the difference between the means of all elements in u and the standard deviation of each element in u; and u i The j-th element u i，j The standardized value of the sample is u i，j with u i The difference of the means of all elements and u i The ratio of the standard deviations of each element in the sample; and the normalization includes global normalization and sample normalization, where: u i The j-th element u i，i The global normalized value is u i，i The difference between u and the minimum value among all elements of u; the ratio of the difference between u and the maximum and minimum values ​​among all elements of u; and u i The j-th element u i，i The sample normalization value is u i，j with u i The difference between the minimum values ​​of all elements in the array, and u i The ratio of the difference between the maximum and minimum values ​​of each element in the array.

[0011] Optionally, in some embodiments of this application, conceptual characterization processing of device operation data includes: inputting the device operation data into a recurrent neural network, and determining its latent space correlation matrix R through the recurrent neural network, wherein, N is the number of hidden layer neurons in the recurrent neural network; and the concept representation matrix C is determined based on the following formula: C = R(R + α) -2 I) -1 ,in, It is an identity matrix, and α is the filtering intensity.

[0012] In some embodiments of this application, the method may optionally further include: initializing the recurrent neural network.

[0013] In some embodiments of this application, optionally, the conceptual representation processing of fault samples generated from fault records includes: inputting each fault sample into the recurrent neural network, and determining the latent space correlation matrix R′ corresponding to each fault sample through the recurrent neural network, wherein, The fault characterization matrix C′ corresponding to each fault sample is determined based on the following formula: C′=R′(R′+α) - 2 I) -1 ; and average the elements in the fault characterization matrix C′ corresponding to each fault sample to generate a fault characterization matrix corresponding to the fault record.

[0014] In some embodiments of this application, the distance may optionally be determined based on the following formula. in, This represents taking the square of the Frobenius norm.

[0015] In some embodiments of this application, the method may optionally further include: performing conceptual characterization processing on the updated fault samples to obtain an updated fault characterization matrix; and determining the distance between the conceptual characterization matrix and the updated fault characterization matrix to determine whether a fault exists.

[0016] In some embodiments of this application, optionally, the conceptual characterization process for the updated fault sample includes: determining the corresponding fault sample u. new Fault characterization matrix in, The updated fault characterization matrix is ​​determined based on the following formula. Where β is the update rate of the fault characterization matrix, and β∈[0,1].

[0017] According to another aspect of this application, a fault diagnosis device for a battery swapping station is provided, the device comprising: a memory configured to store instructions; and a processor configured to execute the instructions to perform any of the methods described above.

[0018] According to another aspect of the present application, there is provided a battery swap station comprising any of the fault diagnosis apparatuses as described above.

[0019] According to another aspect of the present application, there is provided a computer readable storage medium having instructions stored therein, which when executed by a processor, cause the processor to perform any of the methods as described above.

[0020] The battery swap station equipment fault diagnosis method and apparatus, the battery swap station and the storage medium provided by the embodiments of the present application can be used to detect whether the battery swap station equipment has a fault according to log data generated by equipment operation, so as to avoid the expansion of the fault influence. BRIEF DESCRIPTION OF DRAWINGS

[0021] The above and other objects and advantages of the present application will become more complete from the following detailed description of the application taken in conjunction with the accompanying drawings, in which like reference characters refer to like elements throughout the figures, and in which:

[0022] Figure 1 A battery swap station equipment fault diagnosis method according to an embodiment of the present application is shown;

[0023] Figure 2 A recurrent neural network according to an embodiment of the present application is shown;

[0024] Figure 3 A battery swap station equipment fault diagnosis method according to an embodiment of the present application is shown;

[0025] Figure 4 A battery swap station equipment fault diagnosis method according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0026] For brevity and clarity of the description, the principles of the present application are described herein primarily with reference to exemplary embodiments thereof. However, it will be readily apparent to those skilled in the art that the same principles can be equally applied to all types of battery swap station equipment fault diagnosis methods and apparatuses, battery swap stations and storage media, and can be implemented therein, any such changes not departing from the true spirit and scope of the present application.

[0027] According to an aspect of the present application, there is provided a battery swap station equipment fault diagnosis method. As Figure 1As shown, the battery swap station equipment fault diagnosis method 10 (hereinafter referred to as method 10) includes the following steps: in step S102, the equipment operation data is conceptually characterized to obtain a concept characterization matrix; in step S104, the fault samples generated according to the fault records are conceptually characterized to obtain a fault characterization matrix about the fault records; and in step S106, the distance between the concept characterization matrix and the fault characterization matrix is determined, and it is determined whether there is a fault according to the distance. The method 10 can determine whether the equipment of the battery swap station has a fault according to the equipment operation data and the fault records through the above steps, so that the fault can be found at an early stage, and the damage caused by the fault can be avoided from being enlarged.

[0028] In step S102, the method 10 performs concept characterization processing on the equipment operation data to obtain a concept characterization matrix. Concept characterization processing can reduce the difficulty of data processing without losing data features, facilitating comparison with other characterization matrices to determine the similarity between data.

[0029] In some embodiments of the present application, step S102 can specifically include the following process: (1) input the equipment operation data into a recurrent neural network, and determine the hidden space correlation matrix R thereof through the recurrent neural network, wherein, N is the number of hidden layer neurons of the recurrent neural network; (2) determine the concept characterization matrix C based on the following formula: C = R(R + a -2 I) -1 , wherein, and I is an identity matrix, and a is a filtering strength. Extracting key timing features from equipment operation data and converting them into a concept characterization matrix can be used to measure the distance between any two time series in subsequent programs. In some embodiments of the present application, for a recurrent neural network that is first put into application, the method 10 further includes the following steps (not shown in the figure): initializing the recurrent neural network, which will be described in detail below. Figure 1

[0030] (1) First, the weight initialization of the recurrent neural network can be performed. Figure 2 A recurrent neural network is shown in the figure, which includes input layer neurons (a total of S) and hidden layer neurons (a total of N, inside the frame line), the number of input layer neurons S corresponds to the number of variables of the time series, for example, for a single variable time series, the corresponding number of input layer neurons S = 1. The number of hidden layer neurons N represents the dimension of the high-dimensional hidden space. The weight of the recurrent neural network structure connecting the input layer to the hidden layer ​Each element of can be randomly initialized with uniform distribution U(-1, 1). In addition, normal distribution, chi-square distribution, lognormal distribution and other probability distributions can also be used for initialization. On the other hand, the number of hidden layers of the recurrent neural network is not limited to 1 layer, and in some examples, multiple hidden layers can be used. The activation function of the hidden layer neuron can be tanh, relu, leakyrelu, sigmoid and the like.

[0031] The neurons in the hidden layer are sparsely connected, and the weights can be randomly initialized with uniform distribution U(-1, 1), and then the weight matrix W is multiplied by a random variable b to control the connectivity rate of the hidden layer. The random variable b obeys the Bernoulli distribution, which takes the value 1 with the probability p and the value 0 with the probability 1-p, and its expectation can be expressed as the connectivity rate of the hidden layer neuron. Then the scaling of the weight is performed, first obtaining the spectral radius of the weight matrix, that is, the maximum eigenvalue of the absolute value of the weight matrix W ρ(|W|), and then dividing each element of the weight matrix W by ρ(|W|) and multiplying it by the scaling coefficient r∈(0, 1).

[0032] (2) Secondly, the neural network hidden space state correlation matrix R is calculated: after the initialization of the recurrent neural network in step (1), a certain sample time series is input to the input neuron, and the state vector of the high-dimensional hidden space of the neural network at each time t i can be obtained The expression is:

[0033] x(t i )=f(W in u(t i )+Wx(t i-1 ))

[0034] In the formula, is the data of the input neuron at t i , and is the state of the hidden neuron at t i-1 , and f(·) is the activation function of the hidden neuron. It should be noted that the length of the time series T is a variable. Finally, the state matrix of the high-dimensional hidden space of the neural network is obtained X=[x1,x2,...,x N ], where x1, x2,..., x N represent the state vectors of the first to Nth hidden neurons at T times. By calculating the correlation coefficient r ij of the state vectors of any two hidden neurons (i and j), the correlation matrix can be obtained. where the element in the ith row and jth column is r ij .

[0035] (3) Finally, the concept representation matrix C is calculated. In mathematical definition, the concept matrix C is the transformation of the state X in the high-dimensional hidden space, which can be understood as a filter representing the time series dynamic characteristics of the fault samples. The expression of the concept representation matrix C is:

[0036] C = R(R + a -2 I) -1

[0037] wherein, is the unit matrix, is the hidden state correlation matrix, and a is the filtering strength of the time series dynamic characteristics of the fault samples; for a larger a value, the global and macro feature components of the dynamic characteristics will be more significant than the local and detailed feature components; and a smaller a value will make the local and detailed feature components more significant.

[0038] Returning to Figure 1 , the method 10 performs concept representation processing on the fault samples generated according to the fault records in step S104 to obtain a fault representation matrix about the fault records. Using the same concept representation processing as in step S102, the concept representation results of the same type of fault samples, i.e., the fault representation matrix, can be further obtained. The fault representation matrix can not only extract common fault mode key features, but also reduce the bias caused by factors such as the geographical location of the equipment, the season, and the unique working conditions.

[0039] Specifically, in some embodiments of the present application, the concept representation processing on the fault samples generated according to the fault records in step S104 can specifically include the following steps: (1) inputting each fault sample in the fault samples into a recurrent neural network and determining a hidden space correlation matrix R' corresponding to each fault sample through the recurrent neural network, wherein, (2) determining a fault representation matrix C' corresponding to each fault sample based on the following formula: C' = R'(R' + a -2 I) -1 ; and averaging each element in the fault representation matrix C' corresponding to each fault sample to generate a fault representation matrix In addition, in other examples, the method of averaging element by element can not be limited. For example, each sample can be assigned a weight value and weighted average element by element to generate a fault representation matrix

[0040] In step S106, method 10 determines the distance between the conceptual representation matrix and the fault representation matrix, and determines whether a fault exists based on the distance. This distance will be used as a fault characteristic value; the smaller the value, the closer the device is to this type of fault. For example, a fault can be determined to exist when the value is below a certain preset value.

[0041] In some embodiments of this application, the distance can be determined based on the following formula. in, This represents taking the square of the Frobenius norm. In other examples, the distance between the concept representation matrix and the fault representation matrix can also be determined using other methods of measuring matrix distance, such as Euclidean distance, Mahalanobis distance, Minkowsky row distance, etc.

[0042] In some embodiments of this application, the obtained raw data may not always be suitable for direct processing. In such cases, method 10 further includes the following steps ( Figure 1 (Not shown in the image): Preprocessing the time-series data generated by equipment operation to produce equipment operation data. In some examples, preprocessing may involve standardizing or normalizing the time-series data. In other examples, preprocessing may be performed using methods such as Cox-boxing or lognormalization. Since the acquired time-series data is not always suitable for direct processing, the purpose of preprocessing is to obtain data aligned in terms of dimensions and scale, and to maximize the preservation of key features of the sequence pattern, thereby facilitating subsequent calculations.

[0043] In some embodiments of this application, the time series data u = [u1, u2, ..., u...] n That is, a total of n samples of time series data can be collected over a period of time. i Let represent the i-th time series sample in the time series data u, where 1≤i≤n and i∈N+.

[0044] Standardization can be divided into global standardization and sample standardization, and one can be chosen to use during preprocessing. Specifically, when performing global standardization, u i The j-th element u i，j The global standardized value is u i，j The ratio of the difference between the mean of all elements in u and the standard deviation of each element in u. In other words, the result of global standardization of the elements = (elements of each time series array - mean of all samples) / standard deviation of all samples.

[0045] When performing sample standardization, u i The j-th element u i，jThe standardized value of the sample is u i，j with u i The difference of the means of all elements and u i The ratio of the standard deviations of each element in the sample. In other words, the result of sample standardization is: (Elements of the i-th time series sample - Mean of the i-th time series sample) / Standard deviation of the i-th time series sample.

[0046] Normalization can be divided into global normalization and sample normalization, and one can be chosen to use during preprocessing. Specifically, when performing global normalization, u i The j-th element u i，j The global normalized value is u i，j The difference between the minimum value and the maximum value of each element in u, and the ratio between the difference between the maximum value and the minimum value of each element in u. In other words, the result after global normalization of the elements = (the minimum value of each element in the time series array) / (the maximum value of all samples - the minimum value of all samples).

[0047] When performing sample normalization, u i The j-th element u i，j The sample normalization value is u i，j with u i The difference between the minimum values ​​of all elements in the array, and u i The ratio of the difference between the maximum and minimum values ​​of each element in the array. In other words, the result of sample normalization is: (element of the i-th time series array - minimum value of the i-th time series sample) / (maximum value of the i-th time series sample - minimum value of the i-th time series sample).

[0048] In some embodiments of this application, method 10 further includes the following steps ( Figure 1 (Not shown in the text): The updated fault samples are subjected to conceptual characterization processing to obtain the updated fault characterization matrix; and the distance between the conceptual characterization matrix and the updated fault characterization matrix is ​​determined to determine whether a fault exists.

[0049] For example, in some cases, the generated fault characterization matrix can be updated when the update conditions are met. The update conditions may be: (1) significant changes in the equipment operating environment, such as geographical location, season, and operating conditions, leading to changes in the fault mode; (2) the diagnostic model developed based on fault features cannot correctly identify such fault modes, resulting in missed alarms or false alarms. The above two update conditions reflect the situation where fault feature values ​​cannot correctly express the distance relationship between the equipment's runtime sequence data and the fault mode.

[0050] In some embodiments of the present application, the concept representation processing of the updated failure sample comprises: determining a failure representation matrix corresponding to the updated failure sample u new wherein, The updated failure representation matrix is determined based on the following formula wherein, β is the update rate of the failure representation matrix, and β∈[0, 1]. In other examples, the method of updating the failure representation matrix can also be an exponential smoothing method or the like.

[0051] To more clearly describe the working principle of the present application, the following will take Figure 3 and Figure 4 as examples to illustrate the working principle of the battery swap station equipment failure diagnosis method. The following examples do not constitute additional limitations on the independent implementation of other embodiments of the present application.

[0052] As shown in Figure 3 , the battery swap station equipment failure diagnosis method 20 (hereinafter referred to as method 20) comprises the following processes. First, in step S201, the equipment operation time series data is obtained. In step S202, the obtained data is preprocessed to generate equipment operation data. Subsequently, in step S203, the concept representation matrix is extracted by the concept representation module; in step S204, the failure representation matrix is extracted by the failure representation module. In step S205, it is determined whether the update condition is met, if the update condition is met, it enters step S206 and updates the failure representation module to generate an updated failure representation matrix; if the update condition is not met, it directly enters step S207. In step S207, the relationship representation module receives the concept representation matrix and the (updated) failure representation matrix, and in step S208, the distance between the two is calculated as the failure feature value generated by the method 30.

[0053] In Figure 4 ​In the example shown, the working principle of fault diagnosis method 30 (hereinafter referred to as method 30) will be explained by combining feature extraction device 301 and fault recording device 302. As shown in the figure, firstly, feature extraction device 301 acquires and preprocesses device runtime sequence data in step S301 to generate device runtime data. Subsequently, in step S302, the device runtime data is subjected to conceptual representation processing to obtain the conceptual representation matrix of device runtime. In step S303, fault record A from fault recording device 302 is acquired and subjected to fault representation processing to obtain the fault representation matrix. Subsequently, in step S304, feature extraction device 301 detects whether any new fault records are generated. If so, in step S305, newly generated fault records B and C are acquired and the current fault representation matrix is ​​updated; otherwise, the process proceeds directly to step S306. In step S306, method 30 performs relational representation processing on the conceptual representation matrix of device runtime data and the (updated) fault representation matrix to obtain the fault mode distance, and can mark faults with distances less than a threshold as faults.

[0054] According to another aspect of this application, a fault diagnosis device for a battery swapping station is provided, the device comprising: a memory configured to store instructions; and a processor configured to execute instructions to perform any of the methods described above.

[0055] According to another aspect of this application, a battery swapping station is provided, which includes any of the fault diagnosis devices described above.

[0056] According to another aspect of this application, a computer-readable storage medium is provided, wherein instructions are stored, which, when executed by a processor, cause the processor to perform any of the fault diagnosis methods for battery swapping station equipment described above. The computer-readable medium referred to in this application includes various types of computer storage media, and can be any available medium accessible by a general-purpose or special-purpose computer. For example, the computer-readable medium may include RAM, ROM, EPROM, E... 2The computer readable media includes the computer storage medium discussed above. Additionally, or alternatively, the computer readable media can also include a computer readable storage medium other than the computer storage medium discussed above, which other computer readable storage medium is also configured to store program code. The program code stored on one or more of the computer readable storage media is configured to cause a processor to perform one or more of the methods, techniques, or procedures described herein. The program code stored on one or more of the computer readable storage media is configured to cause a processor to perform one or more of the methods, techniques, or procedures described herein. The program code stored on one or more of the computer readable storage media is configured to cause a processor to perform one or more of the methods, techniques, or procedures described herein.

[0057] The method and device for diagnosing faults of battery swap station provided by some embodiments above, the battery swap station, and the storage medium provide a fault detection mechanism based on an unsupervised recurrent neural network and adapted to complex fault modes of the battery swap station, variable-length multivariate time series, a small number of fault samples, and changing fault modes with the environment. The mechanism has the ability to learn the representation of a class, and thus can represent any fault class. In addition, the mechanism can measure the distance relationship between the time series data corresponding to the device and the fault mode, and has boundedness and monotonicity. The mechanism can also adapt to large changes in the device operating environment and operating conditions, update the fault extraction model online, and ensure the accuracy of the fault feature values. The mechanism can be used to detect whether the battery swap station device has a fault according to log data generated by the device, thereby avoiding the expansion of the fault impact.

[0058] The above merely provides a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Those skilled in the art can think of other feasible changes or replacements according to the technical range disclosed in the present application, and such changes or replacements are all included in the protection scope of the present application. The embodiments and features in the embodiments can be combined with each other without conflict. The protection scope of the present application is subject to the description of the claims.

Claims

1. A method for diagnosing equipment faults in a battery swapping station, characterized in that, The method includes: The equipment operation data is processed to obtain a conceptual representation matrix; The fault samples generated from the fault records will be subjected to conceptual characterization processing to obtain a fault characterization matrix for the fault records. Determine the distance between the concept representation matrix and the fault representation matrix, and determine whether a fault exists based on the distance; The updated fault samples are subjected to conceptual characterization processing to obtain the updated fault characterization matrix; and Determine the distance between the concept representation matrix and the updated fault representation matrix to determine whether a fault exists. The conceptual representation of equipment operation data includes: The device operating data is input into a recurrent neural network, and its latent space correlation matrix R is determined through the recurrent neural network, wherein... N is the number of hidden layer neurons in the recurrent neural network; and The concept representation matrix C is determined based on the following formula: C = R(R + ... α -2 I) -1 ,in, It is an identity matrix, and α is the filtering intensity.

2. The method according to claim 1, wherein, The method further includes: preprocessing the time-series data generated by the operation of the equipment to generate the equipment operation data.

3. The method according to claim 2, wherein, The time series data u = [u1, u2, ..., u n ], and u i This represents the i-th time series sample in the time series data u, where 1 ≤ i ≤ n and i ∈ N+. The preprocessing includes standardization or normalization of the time series data, wherein: The standardization includes global standardization and sample standardization, wherein: u i The j-th element u i,j The global standardized value is u i,j The ratio of the difference between the means of all elements in u and the standard deviation of each element in u; and u i The j-th element u i,j The standardized value of the sample is u i,j with u i The difference of the means of all elements and u i The ratio of the standard deviations of the elements in the matrix; and The normalization includes global normalization and sample normalization, wherein: u i The j-th element u i,j The global normalized value is u i,j The difference between the minimum value and the maximum value of u; the ratio of the difference between the maximum value and the minimum value of u; and u i The j-th element u i,j The sample normalization value is u i,j with u i The difference between the minimum values ​​of all elements in the array, and u i The ratio of the difference between the maximum and minimum values ​​of each element in the array.

4. The method according to claim 1, wherein, The method further includes: initializing the recurrent neural network.

5. The method according to claim 1, wherein, The conceptual characterization process for fault samples generated from fault records includes: Each fault sample in the fault samples is input into the recurrent neural network, and the latent space correlation matrix R′ corresponding to each fault sample is determined by the recurrent neural network, where, The fault characterization matrix C′ corresponding to each fault sample is determined based on the following formula: C′=R′(R′+α) -2 I) -1 ;as well as The average value of each element in the fault characterization matrix C′ corresponding to each fault sample is taken to generate the fault characterization matrix corresponding to the fault record.

6. The method according to claim 5, wherein, The distance D is determined based on the following formula: in, This represents taking the square of the Frobenius norm.

7. The method according to claim 1, wherein, Conceptual characterization of the updated fault samples includes: Determine the corresponding fault sample u after the update new Fault characterization matrix in, The updated fault characterization matrix is ​​determined based on the following formula. in, β is the update rate of the fault characterization matrix, and β∈[0,1].

8. A fault diagnosis device for battery swapping station equipment, characterized in that, The device includes: Memory configured to store instructions; and A processor configured to execute the instructions to perform the method as described in any one of claims 1-7.

9. A battery swapping station, characterized in that, The battery swapping station includes the fault diagnosis device as described in claim 8.

10. A computer-readable storage medium storing instructions, characterized in that, When the instruction is executed by the processor, it causes the processor to perform the method as described in any one of claims 1-7.

Images