Charging module state identification method and device based on KHELM-DBO

Through the KHELM-DBO method, the fault characteristics of the charging module are extracted by wavelet packet decomposition and information entropy, combined with the state recognition model, the accuracy and robustness of the state recognition of the DC charging power supply module are improved, and the robustness in the existing technology is solved and the problem of insufficient accuracy.

CN119089336APending Publication Date: 2024-12-06STATE GRID HEBEI ENERGY TECH SERVICE CO LTD +1

Patent Information

Application Number
CN202410785914.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-06-18
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

The existing DC charging power module status recognition methods are insufficiently robust and accurate, making it difficult to effectively monitor charging safety and conversion efficiency.

Method used

Using the KHELM-DBO method, wavelet packet decomposition is performed by obtaining multiple working performance parameter sequences of the charging module, selecting the energy value ratio of the reconstruction node and calculating the main node reconstruction coefficient mean value and wavelet packet information entropy, which is combined into a fault. The feature vector is used and the fault state of the charging module is determined using the trained state recognition model.

Benefits of technology

It improves the robustness and accuracy of charging module status recognition, and ensures the safe operation of the emergency backup power system of the power station.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119089336A_ABST
    Figure CN119089336A_ABST
Patent Text Reader

Abstract

The invention provides a charging module state identification method and device based on KHELM-DBO, and belongs to the technical field of fault diagnosis. The method comprises the following steps: acquiring a plurality of working performance parameter sequences of a charging module; wherein each working performance parameter sequence is a value of one working performance parameter at a plurality of moments; for each working performance parameter sequence, performing wavelet packet decomposition on the working performance parameter sequence to obtain a plurality of reconstruction nodes corresponding to the working performance parameter sequence, and selecting a main node of the working performance parameter sequence from the reconstruction nodes corresponding to the working performance parameter sequence based on an energy value proportion of each reconstruction node; calculating a reconstruction coefficient mean value of each main node and a wavelet packet information entropy of each working performance parameter sequence, and combining the reconstruction coefficient mean value and the wavelet packet information entropy as a fault feature vector of the charging module; and inputting the fault feature vector into a trained state recognition model to obtain a fault state of the charging module. According to the invention, robustness and accuracy of state identification of the charging module can be improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a charging module state identification method and device based on KHELM-DBO. Background Art

[0002] The emergency backup power supply system of the power station uses a DC charging power module to charge the battery. The performance status of the DC charging power module has a significant impact on the battery charging safety, module insulation degradation and conversion efficiency. Therefore, monitoring the performance status of the DC charging power module is an important part of the safe operation of the emergency backup power supply system of the power station.

[0003] The current methods for identifying the status of DC charging power modules include: 1. The analytical model-based method simulates the circuit working state, predicts the voltage or current, and compares it with the reference value; 2. The data-based modeling method is usually built on the modeling framework under offline conditions, and uses expert experience methods to correct the learned model parameters; 3. The signal processing-based method decomposes and transforms the data according to the signal analysis strategy, extracts feature information from it, and compares the change rules of the feature information before and after the fault to achieve fault diagnosis and identification. However, the robustness and accuracy of the above methods still need to be improved. Summary of the invention

[0004] The embodiment of the present invention provides a charging module status recognition method and device based on KHELM-DBO to solve the problem of improving the robustness and accuracy of charging module status recognition.

[0005] In a first aspect, an embodiment of the present invention provides a charging module status identification method based on KHELM-DBO, comprising:

[0006] Acquire multiple working performance parameter sequences of the charging module; wherein each working performance parameter sequence is a value of a working performance parameter at multiple moments;

[0007] For each work performance parameter sequence, wavelet packet decomposition is performed on the work performance parameter sequence to obtain a plurality of reconstruction nodes corresponding to the work performance parameter sequence, and based on the energy value proportion of each reconstruction node, a main node of the work performance parameter sequence is selected from each reconstruction node corresponding to the work performance parameter sequence;

[0008] Calculating the master node reconstruction coefficient mean and wavelet packet information entropy of each work performance parameter sequence, and combining them as the fault feature vector of the charging module;

[0009] The fault feature vector is input into the trained state recognition model to obtain the fault state of the charging module.

[0010] In one possible implementation, the working performance parameter sequence includes a three-phase input current sequence, an output current sequence and an operating temperature sequence, the working performance parameter corresponding to each three-phase input current sequence is one phase of the three-phase input current, and the working performance parameter corresponding to each operating temperature sequence is the operating temperature of a transistor.

[0011] In a possible implementation, before selecting a master node from each reconstruction node corresponding to the work performance parameter sequence based on the energy value proportion of each reconstruction node, the method further includes:

[0012] For each reconstruction node in the work performance parameter sequence, the energy value ratio of the reconstruction node is calculated based on the energy formula; wherein the energy formula is:

[0013]

[0014] Among them, E i is the energy value of the i-th reconstruction node, D i (k) is the wavelet packet coefficient matrix of the i-th reconstruction node, N i is the length of the wavelet packet coefficient matrix of the i-th reconstruction node, I is the number of reconstruction nodes of the working performance parameter sequence, ε i is the energy value ratio of the i-th reconstruction node, and E is the total energy value of each reconstruction node.

[0015] In a possible implementation, calculating the mean of the master node reconstruction coefficient of each work performance parameter sequence includes:

[0016] For each working performance parameter sequence, the reconstruction coefficients of each main node of the working performance parameter sequence are calculated based on the wavelet packet coefficients of each frequency band in the reconstructed sub-signal of the working performance parameter sequence, and the average value of the reconstruction coefficients of each main node is taken as the mean value of the main node reconstruction coefficients of the working performance parameter sequence.

[0017] In a possible implementation, calculating the wavelet packet information entropy of each work performance parameter sequence includes:

[0018] The wavelet packet information entropy of each working performance parameter sequence is calculated based on the wavelet packet information entropy formula; wherein the wavelet packet information entropy formula is:

[0019]

[0020] Where I is the number of working performance parameters in the working performance parameter sequence, ε i is the energy value ratio of the i-th working performance parameter in the working performance parameter sequence.

[0021] In a possible implementation, the master node reconstruction coefficient mean and wavelet packet information entropy of each work performance parameter sequence are calculated and combined as the fault feature vector of the charging module, further comprising:

[0022] Normalizing the mean values ​​of the master node reconstruction coefficients of each work performance parameter sequence to obtain the normalized mean values ​​of the master node reconstruction coefficients of each work performance parameter sequence;

[0023] Normalizing the wavelet packet information entropy of each work performance parameter sequence to obtain the normalized wavelet packet information entropy of each work performance parameter sequence;

[0024] The normalized mean of the main node reconstruction coefficient of each working performance parameter sequence and the normalized wavelet packet information entropy of each working performance parameter sequence are combined as the fault feature vector of the charging module.

[0025] In a possible implementation, a plurality of working performance parameter sequences of the charging module are obtained, including:

[0026] A plurality of working performance parameter sequences of the charging module are obtained in a rolling manner with a preset time period as a cycle.

[0027] In a possible implementation, the state recognition model is a hybrid kernel extreme learning machine.

[0028] In a possible implementation, before the fault feature vector is input into the trained state recognition model, the following is further included:

[0029] Obtain an initial state recognition model;

[0030] The initial state recognition model is trained using a dung beetle optimization algorithm and a training sample set to obtain a trained state recognition model; wherein the training sample set includes a plurality of training samples, each training sample is a group of fault feature vectors, and the label of each training sample is a fault state code.

[0031] In a second aspect, an embodiment of the present invention provides a charging module state identification device based on KHELM-DBO, comprising:

[0032] An acquisition module, used to acquire multiple working performance parameter sequences of the charging module; wherein each working performance parameter sequence is a value of a working performance parameter at multiple moments;

[0033] A reconstruction module is used to perform wavelet packet decomposition on each work performance parameter sequence to obtain a plurality of reconstruction nodes corresponding to the work performance parameter sequence, and select a main node of the work performance parameter sequence from each reconstruction node corresponding to the work performance parameter sequence based on the energy value proportion of each reconstruction node;

[0034] A feature extraction module, used to calculate the mean value of the master node reconstruction coefficient and the wavelet packet information entropy of each working performance parameter sequence, and combine them as the fault feature vector of the charging module;

[0035] The state recognition module is used to input the fault feature vector into the trained state recognition model to obtain the fault state of the charging module.

[0036] The embodiment of the present invention provides a charging module state identification method and device based on KHELM-DBO, which uses multiple working performance parameter sequences of the charging module as information to characterize the operating state of the charging module, selects the main node that almost contains all the information of the original signal through the energy proportion of the reconstructed node, and describes the change trajectory of the main node signal waveform through the mean value of the reconstruction coefficient of the main node, reflects the change of uncertainty and complexity of each working performance parameter sequence in the time and frequency domain through the wavelet packet information entropy, and finally combines to obtain a fault feature vector, and combines the state recognition model to determine the fault state of the charging module, thereby improving the robustness and accuracy of the charging module state recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0038] Figure 1 is an application scenario diagram of a charging module state identification method based on KHELM-DBO provided in an embodiment of the present invention;

[0039] Figure 2 is a flow chart of an implementation method of a charging module state identification method based on KHELM-DBO provided in an embodiment of the present invention;

[0040] Figure 3 is a schematic diagram of an experimental platform provided by an embodiment of the present invention;

[0041] Figure 4 is a schematic diagram of the structure of a simulation circuit provided by an embodiment of the present invention;

[0042] Figure 5A It is a simulation diagram of the operating temperature when an open circuit fault occurs in the MOSFET of the secondary LLC converter switching power supply without protection provided by an embodiment of the present invention;

[0043] Figure 5B It is a simulation diagram of the operating temperature when an open circuit fault occurs in the secondary LLC converter switching power supply MOSFET when there is protection provided by an embodiment of the present invention;

[0044] Figure 5C It is a simulation diagram of three-phase input current and output current after a short circuit fault occurs in a single MOSFET provided by an embodiment of the present invention;

[0045] Fig. 6A is a diagram of fault detection results when S0 and S0* are short-circuited, provided by an embodiment of the present invention;

[0046] Figure 6B is a diagram of VD1 open circuit fault detection results provided by an embodiment of the present invention;

[0047] Figure 6C is a diagram of a D3 short circuit fault detection result provided by an embodiment of the present invention;

[0048] Fig.6D is a diagram of the S1 short circuit fault detection result provided by an embodiment of the present invention;

[0049] Figure 7 is a fault experiment accuracy diagram provided by an embodiment of the present invention;

[0050] Figure 8 It is a structural diagram of a charging module status identification device based on KHELM-DBO provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0051] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present invention. However, it should be clear to those skilled in the art that the present invention may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present invention.

[0052] In order to make the purpose, technical solutions and advantages of the present invention more clear, specific embodiments will be described below in conjunction with the accompanying drawings.

[0053] Charging modules are often used in emergency backup power charging systems of power stations. The health status of charging modules, i.e., the operating status, affects the efficiency and life of emergency backup power. Therefore, it is necessary to identify and locate the faults of charging modules, and at the same time, it is necessary to detect the real-time health status of charging modules to ensure high reliability. At this time, strict requirements and high judgment accuracy are required for the model. The fault data is transformed through improved wavelet packet transform (IWPT), and the relevant parameters of the HKELM algorithm are set, including the number of hidden layer nodes, kernel function type, regularization parameter, etc. The preprocessed input features are mapped to the high-dimensional hidden layer feature space through the hybrid kernel function. The random mapping in the extreme learning machine is replaced by kernel mapping, and the original input data is mapped to a richer feature space, thereby enhancing the expression ability of the model. This method combines the advantages of kernel methods and ELM to achieve more effective and accurate regression and classification tasks. HKELM not only inherits the fast training speed and generalization ability of ELM, but also improves the model's ability to process complex data, thereby achieving comprehensive optimization of model performance; and in order to improve the accuracy and reliability of the model, the dung beetle optimizer (DBO) is used to optimize the regularization coefficient C, mixed kernel parameters σ, m, n, d and weight coefficient λ of the HKELM model. DBO is an optimization algorithm that simulates the behavior of dung beetles in nature and is used to solve global optimization problems. This model is constructed in the detection section of the charging module to detect and analyze the operating status of the charging module in real time, and can achieve high fault identification and positioning accuracy.

[0054] Figure 1 This is an application scenario diagram of the charging module state identification method based on sample entropy provided in an embodiment of the present invention. Figure 1As shown, the main circuit of the charging module includes a rectifier boost front-stage circuit and an electrical isolation current regulation secondary circuit. The front-stage circuit consists of 6 diodes VDi (i=1,2,3,4,5,6), including a 3-p rectifier with a filter capacitor CVD, two MOSFETs (S0 and S0*) with anti-parallel diodes (D0 and D0*) and coupling capacitors (C0 and C0*), a filter capacitor Cb, a diode Db and an inductor Lb for continuous current in the boost circuit. The secondary circuit consists of four MOSFETs (S1 and S1*, S2 and S2*) and anti-parallel diodes (D1 and D1*, D2 and D2*) and coupling capacitors (C1 and C1*, C2 and C2*) in the switching power supply, a resonant capacitor Cr, a resonant inductor Lr and an excitation inductor Lm in the LLC converter, an ideal transformer and two diodes (D3 and D4) and a filter capacitor C3 for a 1-p full-wave rectifier at its output. uA, uB and uC are the three-phase input voltages. iA, iB and iC are the three-phase input currents. Io is the operating current at the module output. Vo is the output voltage, which is equal to the value of the battery. In addition, the input-output ripple Δio (=max(io(k)-min(io(k), k=1,2...n) is calculated, the operating temperature of each MOSFET is monitored (T0 and T0*, T1 and T1*, T2 and T2*), and the output load is tested by increasing the input and output loads under independent working conditions.

[0055] Figure 2 The implementation flow chart of the charging module status identification method based on KHELM-DBO provided in the embodiment of the present invention is described in detail as follows:

[0056] Step 201, obtaining multiple working performance parameter sequences of the charging module; wherein each working performance parameter sequence is the value of a working performance parameter at multiple moments.

[0057] In one possible implementation, the working performance parameter sequence includes a three-phase input current sequence, an output current sequence and an operating temperature sequence, the working performance parameter corresponding to each three-phase input current sequence is one phase of the three-phase input current, and the working performance parameter corresponding to each operating temperature sequence is the operating temperature of a transistor.

[0058] In this embodiment, the selectable working performance parameters include three three-phase input currents, output currents and six MOSFET operating temperatures. One or more of these working performance parameters can be selected according to actual needs, and the corresponding working performance parameter sequence can be obtained. Specifically, the transistor for temperature detection can be selected based on the historical damage data of the transistors at each position, such as the number of damages, and the severity of the impact of the transistor damage on the circuit. In terms of feature extraction, the temperature feature can be selected from the selection of MOSFET to the use in the final circuit, as well as the temperature conditions that can be measured, combined with the data of stable operation and fault experiments.

[0059] Step 202, for each work performance parameter sequence, perform wavelet packet decomposition on the work performance parameter sequence to obtain multiple reconstruction nodes corresponding to the work performance parameter sequence, and select the main node of the work performance parameter sequence from each reconstruction node corresponding to the work performance parameter sequence based on the energy value ratio of each reconstruction node.

[0060] In this embodiment, the process of wavelet packet decomposition is described by taking three-phase input current as an example, and the specific steps are as follows:

[0061] First, the sampled current data is subjected to symmetry analysis and non-zero verification, where the symmetry index λ and the non-zero index σ are expressed as:

[0062]

[0063] In formula (3.1), ψ represents the phase sequence, i.e., phases a, b, and c; N is the total number of sampling points in a sampling window, and the sampling time is T; τ is the fundamental frequency period; N f is the number of sampling points within a fundamental frequency period τ. For the convenience of calculation, T is an integer multiple of τ. k represents the kth fundamental frequency period within the sampling duration T, then k∈[1,T / τ].

[0064] Set a threshold μ. If λ is less than the threshold, the signal is symmetrical and the signal processing is performed using formula (3.2). If λ is greater than the threshold, the signal is asymmetrical and the signal processing is performed using formulas (3.3) to (3.7). If σ is equal to 0, it means that the current is 0 and the signal is maintained at 0 during reconstruction. The final output is the complete reconstructed signal I new The specific signal processing steps are:

[0065] 1) Input a set of sample data I ψ =(I1,I2,…,I N ), the signal symmetry is measured according to formula (3.1). If λ is less than the threshold ρ, the signal is symmetrical, and the reconstructed signal is obtained according to formula (3.2).

[0066]

[0067] In formula (3.2), I k (i) represents the i-th data in the k-th fundamental frequency period of the sampling window, k∈[1,T / τ]; m is the maximum value in the sampling window, that is, m=max(I1,I2,…,I N ); s is the minimum value in the sampling window, that is, s = min(I1, I2, ..., I N );I kmax is the maximum value within the kth fundamental frequency period, i.e., I kmax =max[I k (i)],i∈[1,N f ];I kmin is the minimum value within the kth fundamental frequency period, i.e., I kmin =min[I k (i)]. If λ is less than the threshold ρ, proceed to the subsequent steps.

[0068] 2) If λ is less than the threshold ρ, the signal is asymmetric, and the maximum absolute value of the sampled data in the kth fundamental frequency period is calculated as I' kmax , let I' kmin =-I' kmax , then according to formulas (3.3) to (3.5), the preliminary reconstructed signal I' can be obtained inew .

[0069] I' kmax =max(|I k,i |) (3.3)

[0070] I' kmin =-I' kmax (3.4)

[0071]

[0072] 3) Calculate the original data and the preliminary reconstructed signal I' inew The proportional coefficient C(i) between them.

[0073]

[0074] 4) Combine the symmetric and asymmetric parts of the signal and obtain the final processed signal I according to formula (3.7) new .

[0075]

[0076] Mallat algorithm was proposed by S.Mallat in 1989 and is a fast algorithm for wavelet transform. The fast algorithm for wavelet packet transform decomposition based on Mallat algorithm is expressed as:

[0077]

[0078] In formula (3.8), t is the time series of the input time domain signal, t = 1, 2, ···N, N is the number of sampling points; i = 1, 2, ···, 2 j , j represents the current decomposition level of the wavelet packet; J is the maximum decomposition level of the wavelet packet, J = log2 N ; H, G represent low-frequency and high-frequency filters of wavelet decomposition respectively; is the i-th wavelet packet coefficient of the signal in the j-th layer, where Expressed as the wavelet packet coefficients in the low-frequency part, It is represented as the wavelet packet coefficient in the high frequency part. Input an initial time domain signal f(t), then

[0079] The fast algorithm for wavelet packet transform reconstruction based on Mallat algorithm is:

[0080]

[0081] In formula (3.9), j=J-1,J-2,···,1,0; i=2j,2j-1,···,2,1; h and g are the wavelet reconstruction low-frequency and high-frequency filters respectively.

[0082] The wavelet packet transform based on the Mallat algorithm includes the convolution of the wavelet low-frequency and high-frequency filters with the signal, sub-band sampling, sub-band zero insertion and other key operations. Due to the non-ideal characteristics of the wavelet filter, it is bound to cause repeated information between two adjacent nodes after the wavelet packet decomposition, causing frequency aliasing. If the decomposed signal is further decomposed and reconstructed, the frequency band disorder will be expanded.

[0083] In order to eliminate the frequency aliasing defect caused by the inherent algorithm of wavelet packet transform, this embodiment introduces two operators C and D that can delete redundant information in the frequency band. The goal of operator C is to set the redundant components after the convolution of wavelet decomposition low-frequency filter H and reconstruction low-frequency filter h to zero. Let x represent the low-frequency coefficient of wavelet packet, is the output of operators C and D, then the expression of operator C is:

[0084]

[0085] In formula (3.10), N j is the data length at the current wavelet packet decomposition layer j; n = 0, 1, ···, N j-1 ; k = 0, 1, ···, N j-1 ; Operator V = e -j2π / Nj .

[0086] The goal of operator D is to set the redundant components after the convolution of the wavelet decomposition high-frequency filter G and the reconstruction high-frequency filter g to zero, and its expression is:

[0087]

[0088] In the improved wavelet packet calculation process, the C and D operators are transformed by Fourier transform and its inverse transform, combined with alternate point zero insertion and alternate point sampling to eliminate frequency aliasing defects. The specific algorithm flow is as follows: First, the initial signal After being decomposed by wavelet decomposition filters H and G, that is, Fourier transform is first performed It is converted into a frequency domain signal; then, the operators C and D are used to eliminate the confusing frequency information, and the redundant components after convolution with the wavelet decomposition filters H and G are set to zero, and then the inverse Fourier transform and alternate sampling are performed to obtain and The signal has eliminated the redundant information after decomposition, and it is used as the intermediate signal for interpolation and wavelet reconstruction. The signal reconstructed by h and g is then eliminated by operators C and D to finally obtain the first layer of wavelet packet signal. Similarly, the wavelet packet signal of the first layer is used to obtain the wavelet packet signal of the next layer according to the above process, and so on, to obtain the wavelet packet signal of the specified number of layers.

[0089] This embodiment uses an improved wavelet packet transform based on Mallat algorithm to transform the initial input signal p 1 0 performs three-layer wavelet packet decomposition and reconstruction. The reconstruction nodes are [3,1], [3,2], [3,3], [3,4], [3,5], [3,6], [3,7], [3,8], and the reconstructed sub-signals are p 1 3. p 2 3. p3 3 、p 4 3. p 5 3. p3 6 、p3 7 、p 8 3. The reconstructed node signals and the initial input signals satisfy the following relationship:

[0090] p0 1 =p3 1 +p3 2 +p3 3 +p3 4 +p3 5 +p3 6 +p3 7 +p3 8 (3.12)

[0091] The three-phase current in each fault mode is decomposed using improved wavelet packet transform to obtain the fault characteristics.

[0092] Step 203, calculate the master node reconstruction coefficient mean and wavelet packet information entropy of each work performance parameter sequence, and combine them as the fault feature vector of the charging module.

[0093] In this embodiment, the process of fault feature extraction is as follows: Figure 3 As shown, the specific steps are:

[0094] 1) Select the db3 wavelet basis function, use the improved wavelet packet transform, perform three-layer wavelet packet decomposition and reconstruction on the current in each group of fault samples, and obtain the reconstructed nodes;

[0095] 2) Calculate the energy value of each node and find the proportional coefficient between it and the energy sum of all nodes, and select the node with the highest energy ratio as the main node for fault feature value extraction;

[0096] 3) Calculate the mean value of the main node reconstruction coefficient of each group of sample currents from the wavelet packet coefficients of each frequency band in the reconstructed sub-signal, denoted as M i ;

[0097] 4) Calculate the wavelet packet information entropy of each group of sample currents, denoted as H i ;

[0098] 5) Fusion of the above two feature parameters to construct fault feature vector [M a ,…,M o ,H a ,…,H o ].

[0099] Step 204: input the fault feature vector into the trained state recognition model to obtain the fault state of the charging module.

[0100] In this embodiment, the state recognition model can be a hybrid kernel extreme learning machine (HKELM) regression model. The fault feature vector and the fault state are used as training samples, and the HKELM model is trained using a dung beetle optimizer (DBO) so that the HKELM model learns the connection between the fault feature vector and the fault state, so as to classify the input fault feature vector, determine the corresponding fault state, and realize the fault state recognition of the charging module.

[0101] The embodiment of the present invention uses multiple working performance parameter sequences of the charging module as information to characterize the operating status of the charging module, selects the main node that almost contains all the information of the original signal by reconstructing the energy proportion of the node, and describes the change trajectory of the main node signal waveform by the mean of the reconstruction coefficient of the main node. The change of uncertainty and complexity of each working performance parameter sequence in the time-frequency domain is reflected by the wavelet packet information entropy. Finally, the fault feature vector is combined to obtain the fault state of the charging module, and the state recognition model is combined to determine the fault state of the charging module, thereby improving the robustness and accuracy of the charging module state recognition.

[0102] In a possible implementation, before selecting a master node from each reconstruction node corresponding to the work performance parameter sequence based on the energy value proportion of each reconstruction node, the method further includes:

[0103] For each reconstruction node in the work performance parameter sequence, the energy value ratio of the reconstruction node is calculated based on the energy formula; wherein the energy formula is:

[0104]

[0105] Among them, E i is the energy value of the i-th reconstruction node, D i (k) is the wavelet packet coefficient matrix of the i-th reconstruction node, N i is the length of the wavelet packet coefficient matrix of the i-th reconstruction node, I is the number of reconstruction nodes of the working performance parameter sequence, ε i is the energy value ratio of the i-th reconstruction node, and E is the total energy value of each reconstruction node.

[0106] In this embodiment, the energy spectrum of the ith node after three-layer wavelet packet decomposition is assumed to be E i , the total energy value of the signal is E, and their respective expressions are:

[0107]

[0108] In formula (3.13), D i is the i-th wavelet packet coefficient matrix, k=1,2,···,N i , N i is the length of the coefficient matrix. Let ε i =E i / E, which is expressed as the energy proportion of the i-th node. According to the above formula, the energy proportion of the low-frequency node reaches more than 99%, which almost contains all the information of the original signal, and the detailed information in the original signal is distributed in the remaining node signals.

[0109] Therefore, the low-frequency node is selected as the main node to extract fault features.

[0110] In a possible implementation, calculating the mean of the master node reconstruction coefficient of each work performance parameter sequence includes:

[0111] For each working performance parameter sequence, the reconstruction coefficients of each main node of the working performance parameter sequence are calculated based on the wavelet packet coefficients of each frequency band in the reconstructed sub-signal of the working performance parameter sequence, and the average value of the reconstruction coefficients of each main node is taken as the mean value of the main node reconstruction coefficients of the working performance parameter sequence.

[0112] In this embodiment, by analyzing the main node signal waveform, when different types of faults occur, their change trajectories are different. Therefore, the mean value of the main node wavelet packet reconstruction coefficient is selected as one of the fault characteristic values, denoted as M φ , φ represents the phase sequence.

[0113] One or more master nodes may be selected from each working performance parameter sequence. When there is only one master node, the reconstruction coefficient of the master node is the mean reconstruction coefficient of the master nodes in the working performance parameter sequence. When there are multiple master nodes, the average of the reconstruction coefficients of the master nodes is the mean reconstruction coefficient of the master nodes in the working performance parameter sequence.

[0114] In a possible implementation, calculating the wavelet packet information entropy of each work performance parameter sequence includes:

[0115] The wavelet packet information entropy of each working performance parameter sequence is calculated based on the wavelet packet information entropy formula; wherein the wavelet packet information entropy formula is:

[0116]

[0117] Where I is the number of working performance parameters in the working performance parameter sequence, ε i is the energy value ratio of the i-th working performance parameter in the working performance parameter sequence.

[0118] In this embodiment, in order to deeply mine the detailed information in the signals of each node and discover the abnormal changes in the signals when various faults occur, the concept of information entropy is introduced to realize the extraction of fault features under various fault modes. Information entropy can not only quantitatively describe the energy distribution of time domain signals in the frequency domain, but also reflect the changes in the uncertainty and complexity of the signal in the time and frequency domain. If a signal contains more information and greater uncertainty, the greater the information entropy value. According to the basic theory of information entropy, the wavelet packet information entropy H is defined as:

[0119]

[0120] Therefore, the information entropy of each phase current wavelet packet is selected as another fault characteristic value, denoted as H φ , φ represents the phase sequence.

[0121] In a possible implementation, the master node reconstruction coefficient mean and wavelet packet information entropy of each work performance parameter sequence are calculated and combined as the fault feature vector of the charging module, further comprising:

[0122] Normalizing the mean values ​​of the master node reconstruction coefficients of each work performance parameter sequence to obtain the normalized mean values ​​of the master node reconstruction coefficients of each work performance parameter sequence;

[0123] Normalizing the wavelet packet information entropy of each work performance parameter sequence to obtain the normalized wavelet packet information entropy of each work performance parameter sequence;

[0124] The normalized mean of the main node reconstruction coefficient of each working performance parameter sequence and the normalized wavelet packet information entropy of each working performance parameter sequence are combined as the fault feature vector of the charging module.

[0125] In this embodiment, M φ After normalization, the calculation formula is:

[0126]

[0127] In formula (3.14), M i * is the normalized mean of the low-frequency coefficients, |max(M i )| is the maximum absolute value of the mean of the three-phase low-frequency coefficients.

[0128] Similarly for H φ After normalization, the calculation formula is:

[0129]

[0130] in, is the normalized wavelet packet information entropy, max(H i (x)) is the maximum value of the wavelet packet information entropy.

[0131] Normalization can ensure the consistency and comparability of data, thereby improving the performance and accuracy of the model.

[0132] In a possible implementation, a plurality of working performance parameter sequences of the charging module are obtained, including:

[0133] A plurality of working performance parameter sequences of the charging module are obtained in a rolling manner with a preset time period as a cycle.

[0134] In this embodiment, the same time period can be intercepted for each parameter component sequence. The intercepted time period can be one power frequency cycle, or more complete power frequency cycles. This method can ensure that the intercepted part contains the parameter characteristics of the charging module under a fault state, while shortening the sequence length and reducing the computational complexity of fault feature vector extraction.

[0135] The length of the preset duration should be less than the duration of the working performance parameter sequence, for example, a working performance parameter sequence of 100ms duration is refreshed every 5ms. This embodiment can ensure that the sampling value of the working performance parameter is not missed, and realize real-time and close status recognition of the charging module.

[0136] In a possible implementation, the state recognition model is a hybrid kernel extreme learning machine.

[0137] In this embodiment, ELM is a feedforward neural network learning algorithm, including an input layer, a hidden layer and an output layer. The core idea of ​​the algorithm is to randomly generate hidden layer neurons with constant weights and thresholds during the training process. The ELM regression model is

[0138] f(x)=Hβ (14)

[0139] Where x is the input data vector {xn, n=1,2,...,N}, f(x) is the output of the network, H is the mapping matrix h(x) of the hidden layer, and β is the weight matrix between the hidden layer and the output layer. The mapping matrix h(x) of the ELM model with N input variables and K hidden layers is:

[0140]

[0141] Among them, g(ωk, bk, xn) is a randomly generated activation function, ωk and bk are the weight and threshold of node k in the hidden layer respectively. According to the generalized inverse matrix principle, β can be expressed as follows:

[0142]

[0143] The ELM regression model is

[0144]

[0145] Among them, y is the output data vector {ym, i=1,2,...,M}, C is the penalty coefficient, i0 is the unit diagonal matrix, and HHT is calculated based on the matrix of the random activation function H.

[0146] In order to solve the problem of the uncertainty of the number of hidden layer nodes, Huang et al. constructed a machine learning model KELM using a given kernel function KKELM (xi, xj) instead of random HHT. Generally speaking, the kernel function is a Gaussian function or a polynomial function. Huang et al. further improved KELM, combining Gaussian and polynomial, and proposed a hybrid kernel function KHKELM (xi, xj):

[0147]

[0148] Where σ, m, n, d are the hybrid kernel parameters, λ is the weight coefficient to be optimized, λ∈(0,1). Use the hybrid kernel function matrix KHKELM to replace the random matrix HHT in ELM:

[0149] K HKELM =[K HKELM (x i ,x j )] N×N (19)

[0150] The proposed hybrid kernel extreme learning machine (HKELM) regression model is:

[0151]

[0152] The structure of HKELM neural network is as follows Figure 4 As shown in Figure 2, HKELM combines the advantages of kernel methods and ELM to achieve more efficient and accurate regression and classification tasks. HKELM not only inherits the fast training speed and generalization ability of ELM, but also improves the model's ability to handle complex data, thereby achieving comprehensive optimization of model performance.

[0153] In a possible implementation, before the fault feature vector is input into the trained state recognition model, the following is further included:

[0154] Obtain an initial state recognition model;

[0155] The initial state recognition model is trained using a dung beetle optimization algorithm and a training sample set to obtain a trained state recognition model; wherein the training sample set includes a plurality of training samples, each training sample is a group of fault feature vectors, and the label of each training sample is a fault state code.

[0156] In this embodiment, different types of faults can be set in the simulation circuit, and the corresponding fault feature vectors can be collected to construct a training sample set to train the initial state recognition model so that the state recognition model learns the relationship between the fault feature vector and the fault state. In order to accurately identify the fault type and fault location of the charging module, this application carefully classifies and encodes the fault state of the charging module, with a total of 32 fault types.

[0157] The Dung Beetle Optimizer (DBO) is an optimization algorithm that simulates the behavior of dung beetles in nature and is used to solve global optimization problems. This paper introduces the DBO algorithm and uses the squared mean error as the fitness function to optimize the regularization coefficient C, mixing kernel parameters σ, m, n, d and weight coefficient λ of the HKELM model.

[0158] The objective function of HKELM model optimization based on DBO algorithm is:

[0159]

[0160] Where m is the number of training set samples, fj and yj are the actual output and expected output of the training set samples, respectively.

[0161] The DBO algorithm divides all dung beetle populations into four different subpopulations and simulates five behaviors, namely rolling, dancing, breeding, foraging, and stealing.

[0162] θ represents the rolling ball type dung beetle, which can initially find the direction of foraging based on factors such as the position of the dung beetle, the rolling direction, and environmental changes:

[0163] x θ (t+1)=x θ (t)+α×k×x θ (t-1)+b×|x θ (t)-x θ w | (22)

[0164] Where xθ is the position of the rolling dung beetle. Xθw is the worst individual position in the current population; t is the number of iterations. Α is the natural coefficient, which is -1 when the dung beetle moves away from its initial direction, otherwise it is 1. K is the deflection coefficient, and b is the environmental change coefficient.

[0165] When the ball-rolling dung beetle encounters an obstacle that blocks its rolling path, it needs to reposition itself by dancing to find a new path. The position of the ball-rolling feces is based on the following formula:

[0166] x θ (t+1)=x θ (t)+tan(φ)×(x θ (t)-x θ (t-1)) (23)

[0167] Where φ is the deflection angle,

[0168] η represents reproduction, and the algorithm combines a boundary selection strategy to allow dung beetles to choose a suitable place to lay eggs. μ represents the foraging behavior of dung beetles. The algorithm introduces a strategy to update the boundaries and positions of the best foraging areas to help dung beetle larvae find food sources during foraging. At the same time, λ represents theft. The algorithm updates the strategy by introducing the location of theft, allowing some species to steal food or resources from other species. The position update formulas for the three species are:

[0169]

[0170] Among them, x η 、x μ and x λ are the positions of the feces, ημ and λ populations respectively. The best position in the upper corner represents the current best position of the population. b1 and b2 are unrelated random vectors of the optimization problem dimension. L and U represent the upper and lower bounds of the constraint range of each population, c1~N(0,1) and c2 are random numbers between (0,1), and P is a random vector of normal distribution.

[0171] Finally, according to the above formula, by performing corresponding operations on different types of mantises, the position and fitness of the dung beetle subpopulation are continuously updated until the number of iterations reaches the maximum number of iterations Tmax, and the optimal solution of the HKELM model parameters to be optimized is output.

[0172] In a specific embodiment, in order to verify the effectiveness and feasibility of the fault identification method proposed in this paper, a charging module experimental platform was built, and open circuit fault experiments were carried out on diodes and switches at different positions. The input and output temperature data were collected by an oscilloscope as the training sample set of the DBO-HKELM model, which was divided into a training set and a test set, each accounting for 50% of the total. The main device types used in the platform are: PFC inductor model th6-9021.3, size 940MF, diode model C4D15120D, IGBT model IKW40N120H3, NTC thermistor CWF4B103j3950. The data analysis and processing environment is implemented in computer software.

[0173] In order to improve the robustness of the diagnostic method, the input voltage is set within the interval of ±15% (i.e., 187V~264V) of the rated value (i.e., 220VAC), and the output voltage is set within the interval of ±10% (i.e., 198V~286V) of the rated value (i.e., 220VDC), thereby realizing the simulation of all operating conditions.

[0174] This embodiment takes into account the most common fault types, namely, open circuit fault, short circuit fault and abnormal operating temperature of MOSFET or filter diode in operation. For the charging module, short circuit, overcurrent and overvoltage faults are protected, but the degradation and open circuit fault of MOSFET or rectifier diode have not fully exerted their protection functions. The simulation results of the fault are shown in Figure 1. Figure 5A-Figure 5C As shown in the figure, the operating temperature of the secondary LLC converter switching power MOSFET when there is no protection and an open circuit fault occurs is as follows: Figure 5A As shown in the figure, the operating temperature of the secondary LLC converter switching power MOSFET when an open circuit fault occurs when there is protection is as follows: Figure 5B As shown in Figure 2, the three-phase input current and output current after a single MOSFET short-circuit fault occurs are as follows: Figure 5C shown.

[0175] According to the time domain characteristics of these variables, three risk levels G, Y and R are designed, corresponding to the risk-free level, low risk level and high risk level respectively. Risk level R corresponds to out-of-bounds failure, where a variable exceeds the range determined by the protection value or overload state limit value. Risk level G is that parts and electrical contacts have no or slight deterioration, corresponding to normal operating conditions, and each variable changes within a certain range, generally determined by the rated value and the use reduction factor. The Y level is between G and R, which is usually caused by a gradual deterioration that cannot be ignored. In this embodiment, the risk level threshold of the state characteristics is set, as shown in Table 1.

[0176] Table 1

[0177]

[0178] According to the risk level threshold in Table 1, the fault mode code corresponding to the state characteristics and position of the MOSFET or diode listed in Table 2 can be determined.

[0179] Table 2

[0180]

[0181]

[0182]

[0183] The reconstructed output current under various fault modes is transformed by wavelet packet, and the fault feature vector constructed by the obtained low-frequency coefficient mean and wavelet packet information entropy is shown in Table 3.

[0184] Table 3

[0185]

[0186] The fault feature vector is extracted from the current and temperature of all working conditions under the simulation of the fault at a given time point, and the state recognition model is selected and constructed. The data set is divided, the model is selected and optimized, the model is trained and the model is verified. Finally, the HKELM-DBO model is adopted, and the root mean square error (RMSE) and the mean absolute relative error (MARE) are used for accuracy to obtain the best model parameters. The full-condition unprotected simulated fault state recognition is carried out on the experimental platform, and some results are shown in Figure 6. Fig. 6A This is the fault detection result diagram when S0 and S0* are short-circuited. Figure 6B This is the result diagram of VD1 open circuit fault detection. Figure 6C This is the D3 short circuit fault detection result diagram. Fig.6D The figure is the result of S1 short circuit fault detection. The accuracy of fault diagnosis is statistically analyzed. The statistical results are as follows: Figure 7 As shown, the horizontal axis is the sample number, the vertical axis is the diagnosis result, the red * in the figure is the algorithm diagnosis result, and the blue ○ is the actual fault experiment diagnosis result. The two comparative plots can represent the accuracy of the diagnosis algorithm.

[0187] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0188] The following is an embodiment of the device of the present invention. For details not described in detail therein, reference may be made to the corresponding method embodiment described above.

[0189] Figure 8 The structure diagram of the charging module state identification device based on KHELM-DBO provided in an embodiment of the present invention is shown. For the convenience of explanation, only the part related to the embodiment of the present invention is shown, which is described in detail as follows:

[0190] like Figure 8 As shown, the charging module state identification device 8 based on KHELM-DBO includes:

[0191] An acquisition module 81 is used to acquire multiple working performance parameter sequences of the charging module; wherein each working performance parameter sequence is a value of a working performance parameter at multiple moments;

[0192] The reconstruction module 82 is used to perform wavelet packet decomposition on each work performance parameter sequence to obtain a plurality of reconstruction nodes corresponding to the work performance parameter sequence, and select a main node of the work performance parameter sequence from each reconstruction node corresponding to the work performance parameter sequence based on the energy value proportion of each reconstruction node;

[0193] A feature extraction module 83 is used to calculate the mean value of the master node reconstruction coefficient and the wavelet packet information entropy of each work performance parameter sequence, and combine them as the fault feature vector of the charging module;

[0194] The state recognition module 84 is used to input the fault feature vector into the trained state recognition model to obtain the fault state of the charging module.

[0195] In one possible implementation, the working performance parameter sequence includes a three-phase input current sequence, an output current sequence and an operating temperature sequence, the working performance parameter corresponding to each three-phase input current sequence is one phase of the three-phase input current, and the working performance parameter corresponding to each operating temperature sequence is the operating temperature of a transistor.

[0196] In a possible implementation, the reconstruction module 82 is further configured to:

[0197] Based on the energy value proportion of each reconstruction node, before selecting the main node from each reconstruction node corresponding to the work performance parameter sequence, for each reconstruction node in the work performance parameter sequence, the energy value proportion of the reconstruction node is calculated based on the energy formula; wherein the energy formula is:

[0198]

[0199] Among them, E i is the energy value of the i-th reconstruction node, D i (k) is the wavelet packet coefficient matrix of the i-th reconstruction node, N i is the length of the wavelet packet coefficient matrix of the i-th reconstruction node, I is the number of reconstruction nodes of the working performance parameter sequence, ε i is the energy value ratio of the i-th reconstruction node, and E is the total energy value of each reconstruction node.

[0200] In a possible implementation, the feature extraction module 83 is specifically used to:

[0201] For each working performance parameter sequence, the reconstruction coefficients of each main node of the working performance parameter sequence are calculated based on the wavelet packet coefficients of each frequency band in the reconstructed sub-signal of the working performance parameter sequence, and the average value of the reconstruction coefficients of each main node is taken as the mean value of the main node reconstruction coefficients of the working performance parameter sequence.

[0202] In a possible implementation, the feature extraction module 83 is specifically used to:

[0203] The wavelet packet information entropy of each working performance parameter sequence is calculated based on the wavelet packet information entropy formula; wherein the wavelet packet information entropy formula is:

[0204]

[0205] Where I is the number of working performance parameters in the working performance parameter sequence, ε i is the energy value ratio of the i-th working performance parameter in the working performance parameter sequence.

[0206] In a possible implementation, the feature extraction module 83 is further used to:

[0207] Normalizing the mean values ​​of the master node reconstruction coefficients of each work performance parameter sequence to obtain the normalized mean values ​​of the master node reconstruction coefficients of each work performance parameter sequence;

[0208] Normalizing the wavelet packet information entropy of each work performance parameter sequence to obtain the normalized wavelet packet information entropy of each work performance parameter sequence;

[0209] The normalized mean of the main node reconstruction coefficient of each working performance parameter sequence and the normalized wavelet packet information entropy of each working performance parameter sequence are combined as the fault feature vector of the charging module.

[0210] In a possible implementation, the acquisition module 81 is specifically used for:

[0211] A plurality of working performance parameter sequences of the charging module are obtained in a rolling manner with a preset time period as a cycle.

[0212] In a possible implementation, the state recognition model is a hybrid kernel extreme learning machine.

[0213] In a possible implementation, the state identification module 84 is further configured to:

[0214] Before inputting the fault feature vector into the trained state recognition model, obtaining an initial state recognition model;

[0215] The initial state recognition model is trained using a dung beetle optimization algorithm and a training sample set to obtain a trained state recognition model; wherein the training sample set includes a plurality of training samples, each training sample is a group of fault feature vectors, and the label of each training sample is a fault state code.

[0216] The embodiment of the present invention uses multiple working performance parameter sequences of the charging module as information to characterize the operating status of the charging module, selects the main node that almost contains all the information of the original signal by reconstructing the energy proportion of the node, and describes the change trajectory of the main node signal waveform by the mean of the reconstruction coefficient of the main node. The change of uncertainty and complexity of each working performance parameter sequence in the time-frequency domain is reflected by the wavelet packet information entropy. Finally, the fault feature vector is combined to obtain the fault state of the charging module, and the state recognition model is combined to determine the fault state of the charging module, thereby improving the robustness and accuracy of the charging module state recognition.

[0217] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0218] Those of ordinary skill in the art will appreciate that the templates, units, and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0219] If the module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, it can implement the steps of the above-mentioned various embodiments of the charging module state identification method based on KHELM-DBO. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device that can carry the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory, random access memory, electrical carrier signal, telecommunication signal and software distribution medium.

[0220] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A charging module status identification method based on KHELM-DBO, characterized in that: include: Acquire multiple working performance parameter sequences of the charging module; wherein each working performance parameter sequence is a value of a working performance parameter at multiple moments; For each work performance parameter sequence, wavelet packet decomposition is performed on the work performance parameter sequence to obtain a plurality of reconstruction nodes corresponding to the work performance parameter sequence, and based on the energy value proportion of each reconstruction node, a main node of the work performance parameter sequence is selected from each reconstruction node corresponding to the work performance parameter sequence; Calculating the master node reconstruction coefficient mean and wavelet packet information entropy of each work performance parameter sequence, and combining them as the fault feature vector of the charging module; The fault feature vector is input into a trained state recognition model to obtain the fault state of the charging module.

2. The charging module status identification method based on KHELM-DBO according to claim 1 is characterized in that: The working performance parameter sequence includes a three-phase input current sequence, an output current sequence and an operating temperature sequence. The working performance parameter corresponding to each three-phase input current sequence is one phase of the three-phase input current, and the working performance parameter corresponding to each operating temperature sequence is the operating temperature of a transistor.

3. The charging module status identification method based on KHELM-DBO according to claim 1 is characterized in that: Before selecting a master node from each reconstruction node corresponding to the work performance parameter sequence based on the energy value proportion of each reconstruction node, the method further includes: For each reconstruction node in the work performance parameter sequence, the energy value ratio of the reconstruction node is calculated based on the energy formula; wherein the energy formula is: Among them, E i is the energy value of the i-th reconstruction node, D i (k) is the wavelet packet coefficient matrix of the i-th reconstruction node, N i is the length of the wavelet packet coefficient matrix of the i-th reconstruction node, I is the number of reconstruction nodes of the working performance parameter sequence, ε i is the energy value ratio of the i-th reconstruction node, and E is the total energy value of each reconstruction node.

4. The charging module status identification method based on KHELM-DBO according to claim 1 is characterized in that: The step of calculating the mean of the master node reconstruction coefficients of each work performance parameter sequence includes: For each working performance parameter sequence, the reconstruction coefficients of each main node of the working performance parameter sequence are calculated based on the wavelet packet coefficients of each frequency band in the reconstructed sub-signal of the working performance parameter sequence, and the average value of the reconstruction coefficients of each main node is taken as the mean value of the main node reconstruction coefficients of the working performance parameter sequence.

5. The charging module status identification method based on KHELM-DBO according to claim 1, characterized in that: Calculate the wavelet packet information entropy of each work performance parameter sequence, including: The wavelet packet information entropy of each work performance parameter sequence is calculated based on the wavelet packet information entropy formula; wherein the wavelet packet information entropy formula is: Where I is the number of working performance parameters in the working performance parameter sequence, ε i is the energy value ratio of the i-th working performance parameter in the working performance parameter sequence.

6. The charging module state identification method based on sample entropy according to claim 1, characterized in that: The calculating of the master node reconstruction coefficient mean and wavelet packet information entropy of each work performance parameter sequence and combining them as the fault feature vector of the charging module also includes: Normalizing the mean values ​​of the master node reconstruction coefficients of each work performance parameter sequence to obtain the normalized mean values ​​of the master node reconstruction coefficients of each work performance parameter sequence; Normalizing the wavelet packet information entropy of each work performance parameter sequence to obtain the normalized wavelet packet information entropy of each work performance parameter sequence; The normalized master node reconstruction coefficient mean of each work performance parameter sequence and the normalized wavelet packet information entropy of each work performance parameter sequence are combined as the fault feature vector of the charging module.

7. The charging module status identification method based on KHELM-DBO according to claim 1 is characterized in that: The step of obtaining a plurality of working performance parameter sequences of the charging module includes: A plurality of working performance parameter sequences of the charging module are obtained in a rolling manner with a preset time period as a cycle.

8. The charging module status identification method based on KHELM-DBO according to claim 1, characterized in that: The state recognition model is a hybrid kernel extreme learning machine.

9. The charging module status identification method based on KHELM-DBO according to claim 1, characterized in that: Before inputting the fault feature vector into the trained state recognition model, the method further includes: Obtain an initial state recognition model; The initial state recognition model is trained using a dung beetle optimization algorithm and a training sample set to obtain a trained state recognition model; wherein the training sample set includes multiple training samples, each training sample is a group of fault feature vectors, and the label of each training sample is a fault state code.

10. A charging module status identification device based on KHELM-DBO, characterized in that: include: An acquisition module, used to acquire multiple working performance parameter sequences of the charging module; wherein each working performance parameter sequence is a value of a working performance parameter at multiple moments; A reconstruction module is used to perform wavelet packet decomposition on each work performance parameter sequence to obtain a plurality of reconstruction nodes corresponding to the work performance parameter sequence, and select a main node of the work performance parameter sequence from each reconstruction node corresponding to the work performance parameter sequence based on the energy value proportion of each reconstruction node; A feature extraction module, used to calculate the mean value of the master node reconstruction coefficient and the wavelet packet information entropy of each working performance parameter sequence, and combine them as the fault feature vector of the charging module; The state recognition module is used to input the fault feature vector into a trained state recognition model to obtain the fault state of the charging module.

Citation Information

Patent Citations

  • Microgrid fault diagnosis method for optimizing extreme learning machine based on whale algorithm

    CN112578312A

  • Abnormal current diagnosis method and system based on wavelet decomposition and empirical mode decomposition

    CN112881942A

  • Inverter open-circuit fault identification method based on signal processing reconstruction

    CN115047305A

  • Method for diagnosing analog circuit fault based on vector-valued regularized kernel function approximation

    US20200271720A1

  • Analog-circuit fault diagnosis method based on continuous wavelet analysis and elm network

    US20200300907A1

Cited By

  • SiC MOSFET surge failure real-time discrimination method and system

    CN121348021A