A high-voltage power distribution cabinet fault detection method and system

Abstract

The application discloses a kind of high-voltage distribution cabinet fault detection method and system, it is related to the technical field of fault prediction and health management.The time domain current signal of target distribution cabinet is collected, and the wavelet detail coefficient set is obtained by wavelet transform to time domain current signal;Target wavelet detail coefficient is calculated to obtain the kurtosis after screening to obtain time sequence;According to sliding window, the multiple energy values are obtained by sampling time sequence, if energy value is greater than preset energy threshold, then generate over-limit mark;According to over-limit mark, the energy value set is obtained by continuously sampling the sampling point multiple times, if over-limit mark in energy value set satisfies monotone increasing, then determine that the sampling point is fault point;The inspection coordinates of fault point are sent to inspection equipment, to make inspection equipment collect the temperature of target distribution cabinet and carry out fault early warning.The fault energy is calculated by sliding window, and the fault point is located by threshold and continuous increasing double check, and the temperature is measured by linkage equipment, early warning, anti-electromagnetic interference, and false alarm is reduced.

Description

Technical Field

[0001] This invention relates to the field of fault prediction and health management technology, specifically to a fault detection method and system for high-voltage switchgear. Background Technology

[0002] High-voltage switchgear is widely used in power transmission and distribution networks. Frequent faults, such as aging contacts, surface discharge of insulators, high-resistance grounding, and phase-to-phase short circuits, can easily lead to fires, equipment damage, and other safety accidents, hindering equipment fault prediction and health management technologies. Existing fault detection methods have several inherent shortcomings. Traditional time-limit overcurrent protection uses the power frequency current amplitude for fault identification. However, the fault current amplitude generated by micro-discharge of insulation and high-resistance faults within the switchgear is weak and cannot reach the protection action threshold, resulting in a long-standing problem of fault leakage.

[0003] Publication No. CN121679210A discloses a method for detecting and identifying faults in electrical distribution cabinet wires. The method involves acquiring real-time operating parameter data of multiple wires, identifying faulty wires based on the set of benchmark parameters for normal operation of the wires under test and the relationship between the deviation characteristics of the real-time parameters, and inputting the fault parameter data of the faulty wires into a multi-layer fault identification model. This model contains multiple sequentially connected sub-identification units, each of which detects different types of faults, ultimately obtaining the fault mode.

[0004] Traditional power frequency overcurrent protection is unable to identify weak high-resistance and insulation discharge faults, resulting in missed detections. Conventional fixed frequency band analysis is easily affected by electromagnetic noise interference in the power distribution room, single-point judgment is prone to misjudgment due to instantaneous disturbances, and electrical detection and temperature inspection are independent of each other, making it impossible to quickly locate the fault area. Summary of the Invention

[0005] The purpose of this invention is to solve the problem mentioned in the background art that traditional high-voltage switchgear is easily affected by external signal interference during fault monitoring, resulting in inaccurate fault prediction and health management. Therefore, this invention proposes a fault detection method and system for high-voltage switchgear.

[0006] A first aspect of this invention provides a method for detecting faults in high-voltage switchgear, the method comprising: The time-domain current signal of the target power distribution cabinet is acquired, and the wavelet transform is performed on the time-domain current signal to obtain the wavelet detail coefficient set; The time series sequence is obtained by calculating the kurtosis of the target wavelet detail coefficients and then filtering them; the target wavelet detail coefficients are any one of the wavelet detail coefficients in the set. Obtain a sliding window, sample the time series according to the sliding window to obtain multiple energy values, and generate an over-limit marker if the energy value is greater than a preset energy threshold. The sampling point is sampled multiple times based on the over-limit marker to obtain an energy value set. If the over-limit marker in the energy value set satisfies monotonically increasing, the sampling point is determined to be a fault point. The inspection coordinates of the fault point are sent to the inspection equipment so that the inspection equipment can collect the temperature of the target power distribution cabinet and issue a fault warning.

[0007] Optionally, performing wavelet transform on the time-domain current signal to obtain the wavelet detail coefficient set includes: The current signal type is determined based on the time-domain current signal, the target current signal is decomposed to obtain high-frequency details, and high-frequency detail coefficients are calculated based on the high-frequency details; the current signal types include: A-phase current signal, B-phase current signal, C-phase current signal, and neutral current signal; the target current signal is any one of the time-domain current signals; A high-pass filter is obtained, and the filter coefficients of the high-pass filter are upsampled and expanded to obtain an interpolated high-pass filter. The high-frequency detail coefficients are then subjected to sliding convolution based on the interpolated high-pass filter to obtain convolution features. The normalized convolutional features are used to obtain the target convolutional features, and each target convolutional feature is combined to obtain the wavelet detail coefficient set.

[0008] Optionally, the time series sequences obtained by calculating the kurtosis of the target wavelet detail coefficients and then filtering them include: The frequency range of the target wavelet coefficients is divided into multiple sub-signals by layer by layer using the multi-scale hierarchical logic of the fast kurtosis pyramid. Each sub-signal is substituted into the kurtosis formula to calculate multiple kurtosis values. The kurtosis values ​​are then combined to obtain a kurtosis set. The maximum value of the kurtosis selected from the kurtosis set is recorded as the maximum kurtosis, and the optimal center frequency and bandwidth corresponding to the maximum kurtosis are retained. The maximum kurtosis is then used as a time series.

[0009] Optionally, multiple energy values ​​are obtained by sampling the time series according to the sliding window, and the energy value calculation formula includes:

[0010] in, For the real-time fault energy of phase ψ, K max (m) represents the maximum kurtosis, Δk represents the total number of samples per single power frequency cycle, m represents the original sampling index, k represents the energy sequence index, and Ns represents the total number of sampling points for the entire current signal.

[0011] Optionally, if the out-of-limit markers in the energy value set satisfy a monotonically increasing condition, then the sampling point is determined to be a fault point, including: Identify the out-of-limit markers in the energy set, filter the energy values ​​corresponding to consecutive out-of-limit markers, and if the energy value satisfies monotonically increasing, then determine that the sampling point is a fault point.

[0012] A second aspect of this invention provides a high-voltage switchgear fault detection system, the system comprising: The signal acquisition module is used to acquire the time-domain current signal of the target power distribution cabinet and perform wavelet transform on the time-domain current signal to obtain the wavelet detail coefficient set; The kurtosis determination module is used to calculate the kurtosis of the target wavelet detail coefficients and then filter them to obtain the time series sequence; the target wavelet detail coefficients are any one of the wavelet detail coefficients in the set; The marking and judgment module is used to obtain a sliding window, sample the time series according to the sliding window to obtain multiple energy values, and generate an over-limit marker if the energy value is greater than a preset energy threshold. The fault point determination module is used to continuously sample the sampling point multiple times according to the over-limit marker to obtain an energy value set. If the over-limit marker in the energy value set satisfies monotonically increasing, the sampling point is determined to be a fault point. The fault early warning module is used to send the inspection coordinates of the fault point to the inspection equipment, so that the inspection equipment can collect the temperature of the target power distribution cabinet and issue a fault early warning.

[0013] Optionally, the signal acquisition module includes: The signal decomposition module is used to determine the type of current signal based on the time-domain current signal, decompose the target current signal to obtain high-frequency details, and calculate high-frequency detail coefficients based on the high-frequency details; the types of current signals include: A-phase current signal, B-phase current signal, C-phase current signal, and neutral current signal; the target current signal is any one of the time-domain current signals; The feature extraction module is used to obtain a high-pass filter, upsample and expand the filter coefficients of the high-pass filter to obtain an interpolated high-pass filter, and perform sliding convolution on the high-frequency detail coefficients based on the interpolated high-pass filter to obtain convolution features; The feature normalization module is used to normalize the convolutional features to obtain target convolutional features, and combine each target convolutional feature to obtain a set of wavelet detail coefficients.

[0014] Optionally, the kurtosis determination module includes: The kurtosis calculation module is used to divide the frequency range of the target wavelet coefficients layer by layer into multiple sub-signals through the multi-scale hierarchical logic of the fast kurtosis map pyramid, substitute each sub-signal into the kurtosis formula to calculate multiple kurtosis values, and combine each kurtosis value to obtain a kurtosis set. The kurtosis filtering module is used to compare the maximum value of the kurtosis selection value in the kurtosis set and record it as the maximum kurtosis. The optimal center frequency and bandwidth corresponding to the maximum kurtosis are retained, and the maximum kurtosis is used as a time sequence.

[0015] Optionally, multiple energy values ​​are obtained by sampling the time series according to the sliding window, and the energy value calculation formula includes:

[0016] in, For the real-time fault energy of phase ψ, K max (m) represents the maximum kurtosis, Δk represents the total number of samples per single power frequency cycle, m represents the original sampling index, k represents the energy sequence index, and Ns represents the total number of sampling points for the entire current signal.

[0017] Optionally, the fault warning module is further configured to determine the over-limit markers in the energy set, filter the energy values ​​corresponding to consecutive over-limit markers, and if the energy value satisfies monotonically increasing, then the sampling point is determined to be a fault point.

[0018] The beneficial effects of this invention are: This invention proposes a fault detection method for high-voltage switchgear. It collects the original three-phase and neutral currents within the switchgear and extracts high-frequency fault components through a stationary wavelet transform. The spectral kurtosis is adaptively used to filter and purify the fault impact timing. A sliding window is then used to calculate the stable fault energy. Combined with threshold judgment and continuous monotonically increasing dual verification, the fault sampling point is accurately located. Finally, it is linked with inspection equipment to perform temperature acquisition and fault early warning. This method can completely preserve weak insulation and high-resistance fault characteristics, suppress on-site electromagnetic noise interference, reduce false alarms and missed detections, and simultaneously achieve preliminary electrical fault identification and accurate infrared temperature measurement verification, enabling automatic fault identification and fixed-point inspection. Attached Figure Description

[0019] Figure 1 A flowchart of a high-voltage switchgear fault detection method provided in an embodiment of the present invention. Detailed Implementation

[0020] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0021] This invention provides a method for fault detection in high-voltage switchgear. See also... Figure 1 The method includes the following steps: S101: Collect the time-domain current signal of the target power distribution cabinet, and perform wavelet transform on the time-domain current signal to obtain the wavelet detail coefficient set; S102, calculate the kurtosis of the target wavelet detail coefficients and then filter them to obtain the time series sequence; S103, Obtain a sliding window, sample the time series according to the sliding window to obtain multiple energy values, and generate an over-limit marker if the energy value is greater than the preset energy threshold. S104. Based on the over-limit marker, the sampling point is sampled multiple times to obtain an energy value set. If the over-limit marker in the energy value set satisfies monotonically increasing, the sampling point is determined to be a fault point. S105 sends the inspection coordinates of the fault point to the inspection equipment so that the inspection equipment can collect the temperature of the target distribution cabinet and issue a fault warning.

[0022] The target wavelet detail coefficient is any one of the wavelet detail coefficients in the set; In one implementation, the time-domain current signal of the target distribution cabinet is acquired: the output of the CT of the high-voltage switchgear (i A i B i C i N The original time-domain current is fully acquired without filtering or reduction, resulting in the original current dataset I=[i A i B i C i N Short circuit and insulation discharge faults inside the cabinet only have short-term high-frequency transient components. Deleting the signal would result in the loss of fault characteristics. The instantaneous impact waveform of the fault is fully preserved, providing an original and effective data source for subsequent time-frequency analysis.

[0023] In one implementation, if the energy value is greater than a preset energy threshold, an over-limit marker is generated: the energy value is compared with the preset energy threshold, and if the energy value is greater than the preset energy threshold, an over-limit marker is generated. Distinguish between minor fluctuations in normal load and sudden increases in fault energy, and preliminarily screen out operating conditions without abnormalities; Quickly filter steady-state operating data, retaining only suspected fault samples for secondary verification, thus reducing computational load.

[0024] In one implementation, the original currents of the three phases and neutral line of the high-voltage switchgear are fully acquired, and high-frequency detail components are extracted using stationary wavelet transform. Then, the optimal fault timing sequence is selected by adaptive spectral kurtosis, which fully preserves the high-frequency impact characteristics of insulation discharge and high-resistance faults in the switchgear, eliminates power frequency steady-state load interference, and distinguishes between random electromagnetic noise and fault pulses. High-identity fault features are purified from the original signal, solving the problem of traditional power frequency overcurrent protection failing to detect weak and hidden faults.

[0025] In one implementation, the kurtosis time series is calculated by squared mean based on a single power frequency cycle sliding window to generate a stable quantized energy value. Normal operating conditions are quickly filtered by threshold screening. Then, instantaneous noise disturbances and persistent cabinet faults are distinguished by continuous multi-point monotonically increasing verification. The dual criteria greatly reduce the probability of protection maloperation, take into account the real-time performance of fault detection and the reliability of judgment, and can accurately locate the fault initiation sampling point.

[0026] In one implementation, after the fault location is located by the current signal algorithm, the infrared temperature measurement and inspection equipment is linked to form a two-layer detection system with coarse identification of electrical signals and fine verification of temperature physical characteristics. This system can provide early warning of latent insulation defects and accurately locate the overheating fault area of ​​the high-voltage cabinet. It fully realizes the integrated operation and maintenance capability of automatic fault identification, fixed-point inspection and early warning, and improves the safety management level of the distribution cabinet.

[0027] In one embodiment, performing wavelet transform on a time-domain current signal to obtain a set of wavelet detail coefficients includes: The current signal type is determined based on the time-domain current signal, the target current signal is decomposed to obtain high-frequency details, and the high-frequency detail coefficients are calculated based on the high-frequency details. The current signal types include: phase A current signal, phase B current signal, phase C current signal, and neutral current signal; the target current signal is any one of the time-domain current signals. Obtain a high-pass filter, upsample and expand the filter coefficients of the high-pass filter to obtain an interpolated high-pass filter, and perform sliding convolution on the high-frequency detail coefficients based on the interpolated high-pass filter to obtain convolution features; Normalized convolutional features are used to obtain target convolutional features, and each target convolutional feature is combined to obtain a set of wavelet detail coefficients.

[0028] In one implementation, the current signal type is determined based on the time-domain current signal, ψ∈{A,B,C,N}, representing the discrete time-domain current sampling sequences of phase A, phase B, phase C, and neutral line, respectively; the target current signal is decomposed to obtain high-frequency details: only one level of decomposition is performed (only the first level of high-frequency details is extracted, and low-frequency approximate components are discarded), the stationary wavelet transform (SWT) is not downsampled, and the length of the output sequence after each level of decomposition is equal to the length of the original input sequence.

[0029] In one implementation, the high-frequency detail coefficients are calculated based on the high-frequency details:

[0030] in, Let ψ be the first-order high-frequency detail coefficient of the ψ branch at the k-th sampling point. Here, L is the wavelet normalization scaling factor, L is the wavelet filter length (with a value of 2), and l is the filter sliding index (with a value of 0 < l < L-1). These are the l-th order coefficients of the wavelet high-pass filter. ψ represents the original discrete current sample value of the branch, and k+l-L+1 represents the index of the original current sample position participating in the convolution operation.

[0031] In one implementation, the original length of the standard db2 high-pass filter (L=2) is increased by inserting a zero between the filter coefficients before the first layer decomposition of SWT, resulting in an interpolated high-pass filter adapted to this layer, ensuring that high-frequency timing information is not lost.

[0032] The expanded high-pass filter h ψ Along the original current sequence i ψ Point-by-point sliding convolution: Taking the sampling point k as the output reference, the filter l=0,1 is multiplied by the sampled value of the original current offset position (k+l-L+1) respectively to complete the weighted summation.

[0033] Multiplying the convolution sum by a fixed coefficient This completes wavelet energy normalization to prevent amplitude distortion.

[0034] The first-level decomposition will simultaneously output low-frequency approximation coefficients and high-frequency detail coefficients. The characteristics of short circuits, insulation discharges, and high-resistance faults in the cabinet are all concentrated in the high-frequency range. The low frequency is only the 50Hz steady-state load current, which has no fault information. The low-frequency components are directly discarded, and only the high-frequency detail coefficients are retained to proceed to the next step of fast kurtosis plot calculation.

[0035] In one embodiment, the process of calculating the kurtosis of the target wavelet detail coefficients and then filtering them to obtain the time series sequence includes: The frequency range of the target wavelet coefficients is divided into multiple sub-signals by multi-scale hierarchical logic of fast kurtosis pyramid. Each sub-signal is substituted into the kurtosis formula to calculate multiple kurtosis values. The kurtosis values ​​are combined to obtain a kurtosis set. The maximum value of the kurtosis in the kurtosis set is selected and recorded as the maximum kurtosis. The optimal center frequency and bandwidth corresponding to the maximum kurtosis are retained, and the maximum kurtosis is used as a time series.

[0036] In one implementation, a fast kurtosis graph pyramid multi-scale hierarchical logic is used to divide the frequency range of the target wavelet coefficients into two layers layer by layer to generate a set of narrowband filters with different center frequencies and bandwidths; each filter extracts data from the corresponding frequency band of the target wavelet coefficients to generate an independent narrowband sub-signal. The target wavelet coefficients are a mixed signal across the entire frequency band. The filter bank completes the full-band splitting, providing segmented sub-signals for the calculation of kurtosis of the frequency band. Fault impacts do not have a fixed frequency range, and a fixed single frequency band will lose the weak high-resistance discharge characteristics; only multi-frequency band full coverage can traverse all potential fault characteristic ranges. A single mixed high-frequency signal is split into several independent narrowband sub-signals to achieve frequency domain partitioning analysis.

[0037] In one implementation, multiple kurtosis values ​​are calculated by substituting each sub-signal into the kurtosis formula: traverse the narrowband data corresponding to each group of center frequency and bandwidth, substitute them into the kurtosis formula, calculate the overall kurtosis value of the frequency band, and record the kurtosis corresponding to each group of frequency bands. Each narrowband sub-signal after hierarchical splitting is used as the original sample for kurtosis calculation, quantifying the strength of the impact feature within each frequency band; Kurtosis is a fourth-order statistic that is highly sensitive to instantaneous pulses. Noise-stable signals have extremely low kurtosis and can be quantitatively distinguished from faults and interference. Each frequency range is assigned a quantized impact index, enabling a comparison of the merits and demerits of different frequency bands.

[0038] Kurtosis calculation formula:

[0039] Where K is the kurtosis, M is the total number of sampling points within the currently analyzed narrowband sub-signal, m is the index of the sampling point within the sub-signal (m=1,2,3,...,M), and x m Let m be the signal amplitude at the m-th sampling point. This is the mean of all sampling points of the narrowband sub-signal. Summation of the fourth-order central moments (to amplify the difference in amplitude of the instantaneous impact pulse). Sum the second-order central moments (normalization eliminates kurtosis bias caused by signal amplitude).

[0040] In one implementation, the kurtosis values ​​are combined to obtain a kurtosis set, and the maximum value of the kurtosis values ​​in the kurtosis set is selected and recorded as the maximum kurtosis. For the same sampling time, the kurtosis values ​​of all frequency bands are compared, and the maximum value is selected and recorded as the maximum kurtosis. At the same time, the optimal center frequency and bandwidth corresponding to the maximum value are retained. Multiple frequency band kurtosis values ​​were compared and filtered to eliminate low kurtosis frequency bands dominated by noise, and only the kurtosis of the frequency band with the strongest fault characteristics was retained as the output value. Noise and fault components may coexist at the same time. Only retaining the frequency band with the maximum kurtosis can maximize the suppression of noise and preserve the fault pulse. Each sampling point retains only the optimal fault feature quantization value, significantly reducing the amount of redundant frequency band data.

[0041] In one implementation, the target wavelet detail coefficients corresponding to the A-phase current signal, B-phase current signal, C-phase current signal and neutral current signal are given. The target wavelet detail coefficient is any one of the wavelet detail coefficients in the set, so there are four sets of maximum kurtosis. The four sets of maximum kurtosis are spliced ​​together to obtain the time sequence.

[0042] In one embodiment, multiple energy values ​​are obtained by sampling the time series using a sliding window, and the energy value calculation formula includes:

[0043] in, For the real-time fault energy of phase ψ, K max (m) represents the maximum kurtosis, Δk represents the total number of samples per single power frequency cycle, m represents the original sampling index, k represents the energy sequence index, and Ns represents the total number of sampling points for the entire current signal.

[0044] In one implementation, the real-time fault energy of phase ψ is defined as phase A, phase B, phase C, and neutral. Create a circular buffer of length Δk to store the squared kurtosis values, and calculate the new sample K point by point. max Square the values ​​and store them in the cache. If the window is not full (1≤k≤Δk): accumulate all cached values ​​÷ Δk to get the energy. If the window is full (Δk<≤Ns): remove the oldest data, sum and take the average. Perform independent operations on the four phases to generate four energy sequences.

[0045] In one embodiment, determining a sampling point as a fault point if the out-of-limit markers in the energy value set satisfy a monotonically increasing condition includes: Identify the out-of-limit markers in the energy concentration, filter the energy values ​​corresponding to consecutive out-of-limit markers, and if the energy value satisfies monotonically increasing, then determine that the sampling point is a fault point.

[0046] In one implementation, the energies E(k), E(k-1), E(k-2), and E(k-3) of four consecutive adjacent sampling points are retrieved, and two constraints are simultaneously verified: all energies are greater than a threshold, and the values ​​monotonically increase point by point. Condition met: Mark the first sampling point as the fault initiation point; If the condition is not met: it is determined to be a transient disturbance, and there is no fault output; The system has four independent verification channels: phase A, phase B, phase C, and neutral, which output the phase / ground fault initiation point respectively.

[0047] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention should still fall within the scope of the claims of the present invention.

Claims

1. A method for fault detection in a high-voltage switchgear, characterized in that, The method includes: The time-domain current signal of the target power distribution cabinet is acquired, and the wavelet transform is performed on the time-domain current signal to obtain the wavelet detail coefficient set; The time series sequence is obtained by calculating the kurtosis of the target wavelet detail coefficients and then filtering them; the target wavelet detail coefficients are any one of the wavelet detail coefficients in the set. Obtain a sliding window, sample the time series according to the sliding window to obtain multiple energy values, and generate an over-limit marker if the energy value is greater than a preset energy threshold. The sampling point is sampled multiple times based on the over-limit marker to obtain an energy value set. If the over-limit marker in the energy value set satisfies monotonically increasing, the sampling point is determined to be a fault point. The inspection coordinates of the fault point are sent to the inspection equipment so that the inspection equipment can collect the temperature of the target power distribution cabinet and issue a fault warning.

2. The high-voltage switchgear fault detection method according to claim 1, characterized in that, The wavelet detail coefficient set obtained by performing wavelet transform on the time-domain current signal includes: The current signal type is determined based on the time-domain current signal, the target current signal is decomposed to obtain high-frequency details, and high-frequency detail coefficients are calculated based on the high-frequency details; the current signal types include: A-phase current signal, B-phase current signal, C-phase current signal, and neutral current signal; the target current signal is any one of the time-domain current signals; A high-pass filter is obtained, and the filter coefficients of the high-pass filter are upsampled and expanded to obtain an interpolated high-pass filter. The high-frequency detail coefficients are then subjected to sliding convolution based on the interpolated high-pass filter to obtain convolution features. The normalized convolutional features are used to obtain the target convolutional features, and each target convolutional feature is combined to obtain the wavelet detail coefficient set.

3. The high-voltage switchgear fault detection method according to claim 1, characterized in that, After calculating the kurtosis of the target wavelet detail coefficients and filtering them, the resulting time series sequences include: The frequency range of the target wavelet coefficients is divided into multiple sub-signals by layer by layer using the fast kurtosis pyramid multi-scale hierarchical logic. Each sub-signal is substituted into the kurtosis formula to calculate multiple kurtosis values. The kurtosis values ​​are then combined to obtain a kurtosis set. The maximum value of the kurtosis selected from the kurtosis set is recorded as the maximum kurtosis, and the optimal center frequency and bandwidth corresponding to the maximum kurtosis are retained. The maximum kurtosis is then used as a time series.

4. The high-voltage switchgear fault detection method according to claim 1, characterized in that, Multiple energy values ​​are obtained by sampling the time series using the sliding window, and the energy value calculation formula includes: in, For the real-time fault energy of phase ψ, K max (m) represents the maximum kurtosis, Δk represents the total number of samples per single power frequency cycle, m represents the original sampling index, k represents the energy sequence index, and Ns represents the total number of sampling points for the entire current signal.

5. The high-voltage switchgear fault detection method according to claim 1, characterized in that, If the out-of-limit markers in the energy value set satisfy a monotonically increasing condition, then the sampling point is determined to be a fault point, including: Identify the out-of-limit markers in the energy set, filter the energy values ​​corresponding to consecutive out-of-limit markers, and if the energy value satisfies monotonically increasing, then determine that the sampling point is a fault point.

6. A high-voltage switchgear fault detection system, characterized in that, The system includes: The signal acquisition module is used to acquire the time-domain current signal of the target power distribution cabinet and perform wavelet transform on the time-domain current signal to obtain the wavelet detail coefficient set; The kurtosis determination module is used to calculate the kurtosis of the target wavelet detail coefficients and then filter them to obtain the time series sequence; the target wavelet detail coefficients are any one of the wavelet detail coefficients in the set; The marking and judgment module is used to obtain a sliding window, sample the time series according to the sliding window to obtain multiple energy values, and generate an over-limit marker if the energy value is greater than a preset energy threshold. The fault point determination module is used to continuously sample the sampling point multiple times according to the over-limit marker to obtain an energy value set. If the over-limit marker in the energy value set satisfies monotonically increasing, the sampling point is determined to be a fault point. The fault early warning module is used to send the inspection coordinates of the fault point to the inspection equipment, so that the inspection equipment can collect the temperature of the target power distribution cabinet and issue a fault early warning.

7. A high-voltage switchgear fault detection system according to claim 6, characterized in that, The signal acquisition module includes: The signal decomposition module is used to determine the type of current signal based on the time-domain current signal, decompose the target current signal to obtain high-frequency details, and calculate high-frequency detail coefficients based on the high-frequency details; the types of current signals include: A-phase current signal, B-phase current signal, C-phase current signal, and neutral current signal; the target current signal is any one of the time-domain current signals; The feature extraction module is used to obtain a high-pass filter, upsample and expand the filter coefficients of the high-pass filter to obtain an interpolated high-pass filter, and perform sliding convolution on the high-frequency detail coefficients based on the interpolated high-pass filter to obtain convolution features; The feature normalization module is used to normalize the convolutional features to obtain target convolutional features, and combine each target convolutional feature to obtain a set of wavelet detail coefficients.

8. A high-voltage switchgear fault detection system according to claim 6, characterized in that, The kurtosis determination module includes: The kurtosis calculation module is used to divide the frequency range of the target wavelet coefficients layer by layer into multiple sub-signals through the multi-scale hierarchical logic of the fast kurtosis map pyramid, substitute each sub-signal into the kurtosis formula to calculate multiple kurtosis values, and combine each kurtosis value to obtain a kurtosis set. The kurtosis filtering module is used to compare the maximum value of the kurtosis selection value in the kurtosis set and record it as the maximum kurtosis. The optimal center frequency and bandwidth corresponding to the maximum kurtosis are retained, and the maximum kurtosis is used as a time sequence.

9. A high-voltage switchgear fault detection system according to claim 6, characterized in that, Multiple energy values ​​are obtained by sampling the time series using the sliding window, and the energy value calculation formula includes: in, For the real-time fault energy of phase ψ, K max (m) represents the maximum kurtosis, Δk represents the total number of samples per single power frequency cycle, m represents the original sampling index, k represents the energy sequence index, and Ns represents the total number of sampling points for the entire current signal.

10. A high-voltage switchgear fault detection system according to claim 6, characterized in that, The fault warning module is also used to determine the over-limit markers in the energy concentration, filter the energy values ​​corresponding to consecutive over-limit markers, and if the energy value satisfies monotonically increasing, then the sampling point is determined to be a fault point.

Images