Power supply module fault prediction method and system, and electronic equipment
By applying a combination of xinryl geometric modal decomposition and Transformer model in complex power electronic systems, the problem of low fault prediction accuracy is solved, and higher prediction accuracy and robustness are achieved.
Patent Information
- Application Number
- CN202510214234.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-26
- Publication Date
- 2025-06-13
AI Technical Summary
The problem of low fault prediction accuracy caused by insignificant degradation characteristics and nonlinear time series in complex power electronic systems.
The degraded data is decomposed into multiple sets of inherent modal functions and a set of residual components by using the Transformer model to independently predict each inherent modal function and residual component, and finally the fault prediction result is obtained through fusion reconstruction.
It improves the accuracy and robustness of fault prediction, avoids overfitting and underfitting problems in traditional models, and ensures high prediction accuracy and stability under different frequency conditions.
Smart Images
Figure CN120145106A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of power electronic system fault detection, and particularly to a method, system and electronic device for fault prediction of a power supply module. Background Art
[0002] Power supply modules play a crucial role in electric vehicles, power electronic devices and communication systems, ensuring the normal operation of the devices and the stability of the systems. As the usage time of the devices increases, key components in the power supply modules, such as capacitors, may degrade due to long-term exposure to harsh environments such as high temperature and high voltage, resulting in phenomena such as electrolyte drying and lead corrosion, which in turn affect the overall electrical performance of the module, causing problems such as a decrease in capacitance and an increase in leakage current. In severe cases, these degradations can lead to failures of the power supply module, affecting the normal operation and safety of the devices. Traditional fault detection methods mainly rely on manual inspections and regular tests. Although they can monitor the basic state of the module, due to the long detection cycle and the failure to track the state of the power supply module in real time, early degradation signals are often not detected in a timely manner, resulting in faults being often discovered only after they occur, affecting the reliability of the devices.
[0003] With the development of data-driven technologies, combining real-time monitoring of multi-dimensional data such as voltage, current, and temperature, and using machine learning or deep learning methods to predict faults in power supply modules has become a new solution. Through the analysis of data, potential degradation trends can be identified, and possible faults can be predicted in advance, thereby improving the reliability of the module and the stability of the system. These methods can handle non-linear features and operate stably in a noisy environment, having significant advantages over traditional means. However, despite the great potential of prediction technologies, how to improve the accuracy and robustness of the model under complex working conditions remains a difficult point in current research. Summary of the Invention
[0004] Object of the Invention: To propose a method, system and electronic device for fault prediction of a power supply module, aiming to solve the problem of low fault prediction accuracy caused by unclear degradation characteristics and non-linear time series in complex power electronic systems.
[0005] In the first aspect of the present invention, a method for fault prediction of a power supply module is proposed, and the steps are as follows:
[0006] Build a real physical experiment platform and collect the original degradation data of the power supply module;
[0007] Use symplectic geometric mode decomposition to decompose the original degradation data into multiple groups of intrinsic mode function components and a group of residual components; divide the training set and the test set from the intrinsic mode function components and the residual components;
[0008] Build a Transformer model, train it using the training set, evaluate its performance using the test set, and obtain a fault prediction model;
[0009] Collect the current degradation data of the power supply module to be detected, import the current degradation data into the fault prediction model, and predict the corresponding intrinsic mode function components and residual components of the current fault;
[0010] Fuse and reconstruct the predicted intrinsic mode function components and residual components to obtain the final fault prediction result.
[0011] In a further embodiment of the first aspect, the original degradation data is decomposed into multiple groups of intrinsic mode function components and a group of residual components using symplectic geometric mode decomposition, specifically including:
[0012] Organize the original degradation data into a one-dimensional time series signal according to time;
[0013] Use the power spectral density to adaptively determine the embedding dimension, and reconstruct the one-dimensional time series signal into a multi-dimensional matrix;
[0014] Perform a symplectic geometric matrix similarity transformation on the multi-dimensional matrix to obtain the relevant eigenvalues and eigenvectors of the matrix;
[0015] Use the method of diagonal averaging and similar component recombination to reconstruct the solved eigenvectors to obtain several groups of symplectic geometric components, and each group of symplectic geometric components contains multiple groups of intrinsic mode function components and a group of residual components.
[0016] In a further embodiment of the first aspect, the original degradation data is organized into a one-dimensional time series signal x = x 1 , x 2 ,..., x n ;
[0017] Reconstruct the one-dimensional time series signal x = x 1 , x 2 ,..., x n into a multi-dimensional matrix X:
[0018]
[0019] where m = n - (d - 1)τ, n is the signal length; d is the embedding dimension; τ is the delay time.
[0020] The embedding dimension d is determined by the following formula:
[0021]
[0022] where f s is the sampling frequency, f maxis the maximum main peak frequency in the power spectral density of signal x.
[0023] In a further embodiment of the first aspect, a symplectic geometric matrix similarity transformation is performed on the multi-dimensional matrix, specifically including:
[0024] Construct a Hamiltonian matrix M through the covariance matrix A = X T X:
[0025]
[0026] Let W = M 2 to make the matrix M satisfy the conditions of symplectic geometric similarity transformation, and construct a symplectic orthogonal matrix Q:
[0027]
[0028] In a further embodiment of the first aspect, the relevant eigenvalues and eigenvectors of the matrix are obtained by using the symplectic orthogonal matrix Q, specifically including:
[0029] Set an eigenvector matrix P that satisfies Q = [P 0; 0 P], and construct a coefficient matrix S by using the eigenvector matrix P and the multi-dimensional matrix X:
[0030] S i = P i T X
[0031] Obtain a reconstructed matrix Z by using the eigenvector matrix P and the coefficient matrix S:
[0032] Z i = P i S i
[0033] In the formula, Z i (i = 1, 2,..., d) is the initial single-component reconstructed matrix;
[0034] The reconstructed matrix Z is composed of d groups of initial single-component reconstructed matrices Z i : Z = Z 1 + Z 2 + … + Z d .
[0035] In a further embodiment of the first aspect, the input data of the Transformer model is the symplectic geometric components mapped into vector form, and position encoding is added through sine and cosine functions in the symplectic geometric components mapped into vector form. The formula is as follows:
[0036]
[0037] In the formula, t is the time position index, k is the dimension index, dmodel is the feature extraction dimension of the model, and PE is the position encoding value.
[0038] In a further embodiment of the first aspect, the Transformer model includes a number of encoding layers, and each encoding layer includes a two-layer feed-forward neural network, which processes the input through a linear transformation layer and a ReLU activation function, expressed as:
[0039] FFN(X) = ReLU(xW 3 + b 3 )W 4 + b 4
[0040] where W 3 and W 4 represent the weight matrices of two linear layers respectively; b 3 and b 4 are the corresponding bias terms.
[0041] In the second aspect of the present invention, a power supply module fault prediction system is proposed, and the system includes:
[0042] An acquisition module for acquiring the original degradation data of the target power supply module;
[0043] A decomposition module that uses symplectic geometric mode decomposition to decompose the original degradation data into multiple groups of intrinsic mode function components and a group of residual components;
[0044] A sliding time window partitioning module for partitioning the time series data of each intrinsic mode function component and residual component by a sliding time window, and dividing a training set and a test set therefrom by setting the sliding window size;
[0045] A model training module for constructing a Transformer model, training it with the training set, and evaluating its performance with the test set to obtain a fault prediction model;
[0046] A reconstruction and prediction module for acquiring the current degradation data of the power supply module to be detected, importing the current degradation data into the fault prediction model, and predicting the intrinsic mode function components and residual components corresponding to the current fault; fusing and reconstructing the predicted intrinsic mode function components and residual components to obtain a final fault prediction result.
[0047] In the third aspect of the present invention, an electronic device is proposed, and the electronic device includes: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, the power supply module fault prediction method described in the first aspect is implemented.
[0048] Beneficial effects: The present invention solves the problem of unobvious degraded features in the process of non - linear time - series modeling. Traditional methods often show defects such as low prediction accuracy and poor robustness when dealing with complex signal decomposition and modeling. By introducing the Symplectic Geometric Mode Decomposition (SGMD) technology, the present invention can effectively decompose the degraded signal into multiple Intrinsic Mode Functions and a set of residuals R, thereby enhancing the signal feature extraction ability. Using the Transformer model to independently predict each Intrinsic Mode Function and residual R can more accurately capture the temporal features and non - linear relationships in the signal. This method avoids the over - fitting and under - fitting problems in traditional models, improves the adaptive ability of the model, and ensures high prediction accuracy and stability under different frequency conditions. Description of the Drawings
[0049] Figure 1 This is a flowchart of the power supply module fault prediction method in the embodiment of the present invention. Detailed Embodiments
[0050] In the following description, a large number of specific details are given to provide a more thorough understanding of the present invention. However, it is obvious to those skilled in the art that the present invention can be implemented without one or more of these details. In other examples, some well - known technical features are not described to avoid confusion with the present invention.
[0051] In order to overcome the problem of low prediction accuracy caused by insufficient degraded information of fault feature parameters and difficulty in modeling non - linear time - series of power supply modules under actual working conditions, the present invention combines the Symplectic Geometric Mode Decomposition (SGMD) method and the Transformer prediction model, and proposes a power supply module fault prediction method based on the SGMD - Transformer integrated model. The present invention decomposes the degraded data by using the Symplectic Geometric Mode Decomposition (SGMD) technology, effectively separates different modal components and residual terms in the data, reduces the modeling complexity, and uses the Transformer model to predict the decomposed sequence to complete the modeling and prediction of the fault trend. Finally, based on the SGM - Transformer method, the reconstruction of the predicted values of the decomposed sequence is realized, the trend information of the overall signal is restored, and the accuracy of fault prediction is improved.
[0052] In one embodiment, the present invention proposes a power supply module fault prediction method based on the SGMD - Transformer integrated model. The flowchart of the method is as Figure 1 shown, and includes the following steps:
[0053] Step 1: Collect the original degraded data of the power supply module by building a real - physical experimental platform.
[0054] Step 2: Use symplectic geometric mode decomposition (SGMD) to decompose the collected degradation data into multiple groups of intrinsic mode functions and a group of residuals.
[0055] Step 3: Divide the training set and the test set from the intrinsic mode function components and the residual components in different proportions.
[0056] Step 4: Set the model parameters of the Transformer. Use the training set divided after symplectic geometric mode decomposition (SGMD) to train it, and use the test set to evaluate its performance to obtain the SGMD-Transformer integrated model; collect the current degradation data of the power supply module to be detected, import the current degradation data into the SGMD-Transformer integrated model, and predict the corresponding intrinsic mode function components and residual components of the current fault.
[0057] Step 5: Perform weighted reconstruction on the prediction results of each predicted intrinsic mode function and residual, restore the overall characteristics of the time series, realize the fault prediction of the power supply module, and evaluate the reconstructed prediction results in combination with the root mean square error (RMSE) and the mean absolute error (MAE) to ensure that the fault prediction model under different working conditions has high precision and robustness, thereby effectively improving the reliability of fault prediction.
[0058] As a preferred solution, the specific process of the symplectic geometric mode decomposition (SGMD) method in Step 2 is as follows:
[0059] Step 2.1: Trajectory matrix construction
[0060] Assume that the time series signal x = x 1 , x 2 ,..., x n . According to the Takens embedding theorem, one-dimensional data can be reconstructed into multi-dimensional signals, and then the trajectory matrix X can be constructed as
[0061]
[0062] where m = n - (d - 1)τ. n is the signal length, determined by the time series; d is the embedding dimension; τ is the delay time, generally τ = 1. The embedding dimension d is determined by the following formula:
[0063]
[0064] where f s is the sampling frequency, and f max is the maximum main peak frequency in the power spectral density of the signal x.
[0065] Step 2.2: Symplectic geometric matrix transformation
[0066] Through the covariance matrix \(A = X\) T X, construct the Hamiltonian matrix \(M\):
[0067]
[0068] To make the matrix \(M\) satisfy the conditions of symplectic geometric similarity transformation, let \(W = M\) 2 , construct the symplectic orthogonal matrix \(Q\), and we can get:
[0069]
[0070] Wherein, \(Q\) can be obtained by symplectic similarity transformation.
[0071] Set a Householder matrix \(P\), and construct the matrix \(H=[P\ 0;0\ P]\), it can be proved that \(H\) is also a Householder matrix. Therefore, the matrix \(Q\) can be replaced by the matrix \(H\).
[0072] It can be known that there is a certain correlation between the eigenvalues of matrix \(A\) and matrix \(B\). Assume is the eigenvalue of matrix \(A\), then \(\lambda\) 1 , \(\lambda\) 2 ,..., \(\lambda\) d are the eigenvalues of matrix \(B\), \(P\) i \((i = 1,2,\cdots,d)\) is the eigenvector corresponding to the eigenvalue of matrix \(A\) 2 , and at this time, the problem of the unknown matrix \(B\) can be transformed into the problem of matrix \(A\).
[0073] Through symplectic geometric similarity transformation, the matrix \(P\) preserves the essential characteristics of the original signal. Using the PR decomposition to decompose the matrix \(A\) 2 , the eigenvector matrix \(P\) can be obtained.
[0074] Use the eigenvector matrix \(P\) and the original trajectory matrix \(X\) to construct the coefficient matrix \(S\):
[0075] \(S\) i =P i T \(X\)
[0076] Furthermore, use the eigenvector matrix \(P\) and the coefficient matrix \(S\) to obtain the reconstruction matrix \(Z\):
[0077] \(Z\) i =P i \(S\) i
[0078] Wherein, \(Z\) i \((i = 1,2,\cdots,d)\) is also called the initial single-component reconstruction matrix.
[0079] At this time, the reconstruction matrix \(Z\) consists of \(d\) groups of initial single-component reconstruction matrices \(Z\) iComposition: Z = Z 1 + Z 2 + … + Z d .
[0080] Step 2.3, Diagonal Averaging
[0081] For the reconstruction matrix Z with the feature dimension of m × d i perform diagonal averaging, and we can obtain
[0082]
[0083] According to the above formula, the reconstruction matrix Z can be transformed into d groups of reconstruction independent components Y with the length of n i (y 1 , y 2 , …, y n ). The original time series can be represented by the sum of these d groups of reconstruction components:
[0084] Y = Y 1 + Y 2 + Y 3 + … + Y d
[0085] Step 2.4, Similar Component Recombination
[0086] The decomposition result of the symplectic geometric mode decomposition (SGMD) method may contain components with the same frequency or correlation. Therefore, it is necessary to perform similarity analysis on the reconstructed time series Y and merge the similar components. Since the eigenvectors are sorted from large to small during the decomposition process, the front components usually include the main frequency components of the signal. Starting from Y 1 , compare the subsequent components with Y 1 , select the components with higher similarity to Y 1 and Y 1 to merge and obtain a part of the symplectic geometric component SGC 1 . Then delete the components that make up SGC 1 from the matrix Y to get the residual signal g 1 . Then calculate the normalized mean square error (NMSE), and the formula is as follows:
[0087]
[0088] where h is the number of iterations, and g h represents the residual component. Repeat the above iterative process until the NMSE is less than the given threshold, then the iteration stops, and the final result is obtained:
[0089]
[0090] where N is the number of SGC components obtained, and g(N+1) (n) is the residual component.
[0091] As a preferred solution, the specific setting steps of the Transformer model in step 4 include:
[0092] Step 4.1: Extract features from the input data and map them into vector form, called the input vector. To introduce the time position information of the sequence, positional encoding is used. The positional encoding is generated by sine and cosine functions, and the formula is as follows:
[0093]
[0094] In the formula, t is the time position index, k is the dimension index, d model is the feature extraction dimension of the model, and PE is the positional encoding value.
[0095] Step 4.2: Self-attention mechanism
[0096] X·[W q ,W k ,W v = [Q, K, V]
[0097]
[0098] In the formula, W q , W k , W v are linear transformation matrices used to map the input data X to the query vector Q, the key vector K, and the value vector V. Softmax is the exponential normalization function, and d k is the number of dimensions of K.
[0099] The multi-head self-attention mechanism performs h groups of calculations on the input quantity on this basis, and the calculation formula is as follows:
[0100] A i = A(Q i , K i , V i )
[0101] A MH (X) = W mh ·[A 1 , …, A h
[0102] In the formula, i is the self-attention head number; A i is the self-attention expression of the i-th number; Q i , K i , V iThree groups of mappings for the i-th self-attention head of X, which are calculated in the same way as the single-head self-attention mechanism; W mh is the multi-head parameter matrix; h is the number of self-attention heads; A MH (X) is the multi-head self-attention expression of X.
[0103] Step 4.3: Each Transformer encoding layer also includes a two-layer feed-forward neural network, which processes the input through a linear transformation layer and a ReLU activation function, and can be expressed as:
[0104] FFN(X) = ReLU(xW 3 + b 3 )W 4 + b 4
[0105] In the formula, W 3 and W 4 represent the weight matrices of the two linear layers respectively; b 3 and b 4 are the corresponding bias terms.
[0106] Step 4.4: Each sub-layer adopts a residual connection method to ensure that the input and output are superimposed for the next calculation.
[0107] At the same time, the normalization operation can improve the model stability, and the formula is:
[0108]
[0109] In the formula, μ and σ are the mean and standard deviation of the input, and γ and β are learnable parameters.
[0110] Step 4.5: The output layer maps the features to the target space through a fully connected network, and combines the softmax function to generate a normalized probability distribution for the final prediction.
[0111] In another embodiment, a power supply module fault prediction system based on the SGMD-Transformer integrated model is proposed. The system includes an acquisition module, a decomposition module, a sliding time window division module, a model training module, and a reconstruction and prediction module.
[0112] The acquisition module is used to acquire the degradation data of the power supply module.
[0113] The decomposition module uses the symplectic geometric mode decomposition (SGMD) technology to decompose the data into several intrinsic mode functions and a residual component.
[0114] The sliding time window division module is used to perform sliding time window division on the time series data of each intrinsic mode function and residual component. By setting the sliding window size, the module constructs a training set and a test set to ensure that the input data can reflect the historical trend of the time series.
[0115] The model training module uses the Transformer model to independently train and predict for each intrinsic mode function and residual component.
[0116] The reconstruction and prediction module performs weighted reconstruction on the prediction results of all intrinsic mode functions and residual components to restore the overall time series data for completing the fault prediction of the power supply module. This module calculates the overall error index in combination with the predicted values and generates the final prediction result.
[0117] The implementation methods of each module and module functions in the system are exactly the same as the steps of the method in Embodiment 1, so they will not be elaborated here.
[0118] In another embodiment, the present invention proposes an electronic device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the fault prediction method for the power supply module based on the SGMD-Transformer integrated model as described in Embodiment 1.
[0119] In another embodiment, the present invention proposes a computer-readable storage medium storing a computer program, and the computer program causes a computer to execute the fault prediction method for the power supply module based on the SGMD-Transformer integrated model as described in Embodiment 1.
[0120] In the embodiments disclosed in the present application, the computer storage medium can be a tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. The computer storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of the computer storage medium would include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.
[0121] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed in this application can be implemented by electronic hardware or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods for each specific application to implement the described functions, but such implementation should not be considered to exceed the scope of this application.
[0122] As described above, although the present invention has been shown and described with reference to specific preferred embodiments, it should not be construed as a limitation of the present invention itself. Various changes can be made in its form and details without departing from the spirit and scope of the present invention defined by the appended claims.
Claims
1. A method for predicting power supply module failure, characterized in that: The steps include: Build a real physical experimental platform to collect the original degradation data of the target power supply module; Decomposing the original degradation data into a plurality of groups of intrinsic mode function components and a group of residual components using symplectic geometric mode decomposition; summarizing the intrinsic mode function components and the residual components, and dividing the training set and the test set therefrom according to a predetermined ratio; Based on the Transformer model, the training set and the test set after symplectic geometric mode decomposition are used to train and evaluate it to obtain a fault prediction model; Collecting current degradation data of the power supply module to be detected, importing the current degradation data into the fault prediction model, and predicting the intrinsic mode function component and residual component corresponding to the current degradation data; The predicted intrinsic mode function components and residual components are fused and reconstructed to obtain the final fault prediction result.
2. The power supply module fault prediction method according to claim 1, characterized in that: The original degradation data is decomposed into multiple groups of intrinsic mode function components and a group of residual components using symplectic geometric mode decomposition, specifically including: Arranging the original degradation data into a one-dimensional time series signal according to time; Adaptively determine the embedding dimension using power spectral density, and reconstruct the one-dimensional time series signal into a multi-dimensional matrix; Performing a symplectic geometric matrix similarity transformation on the multidimensional matrix to obtain relevant eigenvalues and eigenvectors of the matrix; The solved eigenvectors are reconstructed using the diagonal average and similar component recombination methods to obtain several groups of symplectic geometric components, each of which contains multiple groups of intrinsic mode function components and a group of residual components.
3. The power supply module fault prediction method according to claim 2, characterized in that: The original degradation data is sorted into a one-dimensional time series signal x=x1,x2,...,x n ; The one-dimensional time series signal x=x1,x2,...,x n Reshape into a multidimensional matrix X: Wherein, m=n-(d-1)τ, n is the signal length; d is the embedding dimension; τ is the delay time.
4. The power supply module fault prediction method according to claim 3, characterized in that: The embedding dimension d is determined by the following formula: In the formula, f s is the sampling frequency, f max is the maximum main peak frequency in the power spectrum density of signal x.
5. The power supply module fault prediction method according to claim 2, characterized in that: Performing a symplectic geometric matrix similarity transformation on the multidimensional matrix specifically includes: Through the covariance matrix A = X T X, construct the Hamiltonian matrix M: Let W = M 2 , so that the matrix M satisfies the conditions of symplectic geometric similarity transformation and constructs the symplectic orthogonal matrix Q:
6. The power supply module fault prediction method according to claim 5, characterized in that: The symplectic orthogonal matrix Q transformation is used to obtain the relevant eigenvalues and eigenvectors of the matrix, including: Set up an eigenvector matrix P, satisfying Q = [P 0; 0P], and use the eigenvector matrix P and the multidimensional matrix X to construct the coefficient matrix S: The reconstruction matrix Z is obtained using the eigenvector matrix P and the coefficient matrix S: Z i =P i S i In the formula, Z i (i=1,2,...,d) is the initial single-component reconstruction matrix; The reconstruction matrix Z is composed of d sets of initial single-component reconstruction matrices Z i Composition: Z = Z1 + Z2 + ... + Z d .
7. The power supply module fault prediction method according to claim 2, characterized in that: The input data of the Transformer model is the symplectic geometric components mapped into vector form, and position encoding is added to the symplectic geometric components mapped into vector form through sine and cosine functions, and the formula is as follows: In the formula, t is the time position index, k is the dimension index, and d model is the feature extraction dimension of the model, and PE is the position encoding value.
8. The power supply module fault prediction method according to claim 1, characterized in that: The Transformer model contains several encoding layers, each of which includes a two-layer feedforward neural network that processes input through a linear transformation layer and a ReLU activation function, expressed as: FFN(X)=ReLU(xW3+b3)W4+b4 Where W3 and W4 represent the weight matrices of the two linear layers respectively; b3 and b4 are the corresponding bias terms.
9. A power supply module fault prediction system, characterized in that: include: An acquisition module, used for acquiring original degradation data of a target power supply module; A decomposition module, using symplectic geometric mode decomposition to decompose the original degradation data into multiple groups of intrinsic mode function components and a group of residual components; A sliding time window partitioning module is used to perform sliding time window partitioning on the time series data of each intrinsic mode function component and residual component, and to divide the training set and the test set from the data by setting the sliding window size; The model training module is based on the Transformer model. It uses the training set to train it and the test set to evaluate its performance to obtain a fault prediction model. The reconstruction and prediction module is used to collect the current degradation data of the power supply module to be detected, import the current degradation data into the fault prediction model, predict the inherent modal function components and residual components corresponding to the current degradation data; fuse and reconstruct the predicted inherent modal function components and residual components to obtain the final fault prediction result.
10. An electronic device, characterized in that: The device comprises: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, the power supply module fault prediction method according to any one of claims 1 to 8 is implemented.