Power grid partial discharge detection method

Abstract

The invention provides a power grid partial discharge detection method, which comprises the steps of detecting a partial discharge signal through a preset unmanned aerial vehicle cluster, and correcting the collected partial discharge signal through motion compensation phase correction; performing hierarchical processing on the partial discharge signal through a multi-scale dynamic convolution network, capturing different frequency band characteristics through a dynamically adjusted convolution kernel group, and performing nonlinear suppression on high-frequency noise through an adaptive wavelet threshold function; extracting feature vectors of a time domain and a frequency domain of the partial discharge signal, and performing discharge type identification according to a fusion result of the feature vectors of the time domain and the frequency domain; and according to time difference positioning and energy gradient attenuation in the unmanned aerial vehicle cluster, accurate coordinate positioning of a discharge source is determined in a three-dimensional space, and a final power grid partial discharge detection result is obtained. According to the invention, accurate detection of partial discharge carried by the unmanned aerial vehicle is realized, the problem of partial discharge detection in a complex electromagnetic environment is solved, and the detection precision and the positioning accuracy are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid partial discharge detection, and particularly to a method for detecting power grid partial discharge. Background Art

[0002] In the aspect of partial discharge detection, in the existing technology, it is difficult to effectively extract partial discharge signals in a complex electromagnetic environment, and it is vulnerable to noise interference, resulting in a reduction in detection accuracy. Some traditional detection methods cannot be applied to the detection method based on unmanned aerial vehicles (UAVs) at the same time, especially the influence of UAV movement on detection, and in particular, the positioning accuracy cannot meet the actual requirements. Summary of the Invention

[0003] The purpose of the present invention is to propose a method for detecting power grid partial discharge, and solve the technical problem of how to achieve high-precision detection of power grid partial discharge in a complex electromagnetic environment based on a UAV platform.

[0004] On the one hand, a method for detecting power grid partial discharge is provided, including:

[0005] Detecting partial discharge signals through a preset UAV cluster, and correcting the collected partial discharge signals through motion compensation phase correction;

[0006] Performing hierarchical processing on the partial discharge signals through a multi-scale dynamic convolution network, capturing features of different frequency bands through a dynamically adjusted convolution kernel group, and nonlinearly suppressing high-frequency noise through an adaptive wavelet threshold function;

[0007] Extracting the feature vectors in the time domain and frequency domain of the partial discharge signals, and identifying the discharge type according to the fusion result of the feature vectors in the time domain and frequency domain;

[0008] Determining the accurate coordinate positioning of the discharge source in three-dimensional space according to the time difference positioning and energy gradient attenuation within the UAV cluster, and obtaining the final power grid partial discharge detection result.

[0009] Preferably, it further includes performing motion compensation phase correction according to the following formula,

[0010]

[0011] where S raw (t) is the original sensor signal; Δd is the change in UAV displacement; λ is the signal wavelength.

[0012] Preferably, the convolution kernel group includes,

[0013]

[0014] where W iis the base convolution kernel; σ is the Sigmoid activation function; α i , β i are learnable weight parameters; s is the input signal feature.

[0015] Preferably, the adaptive wavelet threshold function includes

[0016]

[0017] where λ adapt is the dynamically adjusted threshold; N is the signal length; E b is the background noise energy.

[0018] Preferably, the adaptive wavelet threshold function further includes adjusting the wavelet coefficients according to the following rules

[0019]

[0020] where wj is the wavelet coefficient; σ n is the noise standard deviation.

[0021] Preferably, the non-linear suppression of high-frequency noise includes performing multi-layer wavelet packet decomposition on the partial discharge signal, and applying a dynamic convolution kernel group to the multi-layer high-frequency coefficients, and fusing features through the following cross-scale attention mechanism:

[0022]

[0023] where Q, K, V are the query, key, and value matrices from different wavelet decomposition layers respectively; d k is the dimension of the key vector.

[0024] Preferably, it further includes fusing the time-domain and frequency-domain feature vectors according to the following formula

[0025] F = [TKEO(s(t)), HHT(f), Re(WVD(t,f))]

[0026] where F is the joint feature vector; TKEO is the Teager-Kaiser energy operator; HHT is the marginal spectrum of the Hilbert-Huang transform; WVD is the real part of the Wigner-Ville distribution; s(t) is the original time-domain signal; t is the time variable; f is the frequency variable.

[0027] Preferably, it further includes identifying the discharge type according to the following formula

[0028]

[0029] where is the predicted discharge type category label; SVM(·) is the support vector machine classifier; RF embedding is the feature embedding function of the random forest; F is the time-frequency domain joint feature vector.

[0030] Preferably, the time difference positioning and energy gradient attenuation include performing time difference positioning according to the following formula:

[0031]

[0032] where, Δt ij is the time difference of arrival of signals between UAVs i and j; r i , r j are the three-dimensional space coordinates of UAVs i and j; P is the three-dimensional coordinate of the discharge source to be located; v is the signal propagation speed;

[0033] Perform energy gradient attenuation constraint according to the following formula:

[0034]

[0035] where, is the energy gradient; E k is the energy of the discharge signal received by the k-th sensor; r k is the three-dimensional coordinate of the k-th sensor; P is the three-dimensional coordinate of the discharge source to be solved.

[0036] Preferably, the accurate coordinate positioning of the discharge source includes performing positioning by combining time difference positioning and energy gradient attenuation according to the following formula:

[0037]

[0038] where, P is the three-dimensional coordinate of the discharge source to be solved; Δt ij is the time difference of arrival of signals between UAVs i and j; is the square of the L2 norm of the energy gradient.

[0039] In summary, implementing the embodiments of the present invention has the following beneficial effects:

[0040] The method for detecting partial discharge in the power grid provided by the present invention realizes precise detection of partial discharge carried by UAVs through constructing a UAV collaborative sensing network, performing motion compensation preprocessing, applying a multi-scale dynamic convolution denoising algorithm, dual-domain feature fusion recognition, and time difference-energy gradient joint positioning technology, solves the problem of partial discharge detection in a complex electromagnetic environment, and improves the detection accuracy and positioning accuracy. Description of the Drawings

[0041] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, obtaining other accompanying drawings based on these drawings still belongs to the scope of the present invention.

[0042] Figure 1 It is a schematic diagram of the main process of a method for detecting partial discharge in a power grid in an embodiment of the present invention. Specific embodiments

[0043] To make the purpose, technical solutions and advantages of the present invention clearer, the following will further describe the present invention in detail with reference to the accompanying drawings.

[0044] As Figure 1 shown, it is a schematic diagram of an embodiment of a method for detecting partial discharge in a power grid provided by the present invention. In this embodiment, it is necessary to construct an unmanned aerial vehicle (UAV) collaborative sensing network: deploy a cluster of 3 - 5 UAVs equipped with multi - modal sensors. Each UAV is equipped with a UHF sensor for detecting the UHF signals generated by partial discharge, a broadband acoustic wave array for receiving the acoustic signals of partial discharge, and a nine - axis inertial measurement unit for obtaining the motion attitude information of the UAV.

[0045] Implement multi - machine data synchronization through a distributed topology protocol to ensure that the multi - machine data synchronization error < 2μs, satisfying:

[0046]

[0047] In the formula, Δt sync is the maximum time synchronization error between multi - UAV sensors;

[0048] f max is the highest sampling frequency of the sensor (for example, 500 kHz);

[0049] This formula ensures that the time error of multi - machine data acquisition does not exceed 1 / 10 of the sampling period.

[0050] The method for detecting partial discharge in the power grid includes the following steps:

[0051] Step S1, detect partial discharge signals through a preset UAV cluster, and correct the collected partial discharge signals through motion - compensation phase correction; use IMU data to eliminate the influence of UAV jitter on the sensor, and correct the collected signals through the motion - compensation phase - correction formula.

[0052] In one embodiment, perform motion - compensation phase correction according to the following formula, and compensate for the phase shift caused by UAV jitter through a complex exponential term:

[0053]

[0054] Among them, S raw (t) is the original sensor signal; Δd is the displacement change of the UAV (measured by the IMU); λ is the signal wavelength.

[0055] Meanwhile, establish the relationship matrix between the motion state and the signal phase to achieve three-dimensional space compensation:

[0056]

[0057] In the formula: θ x is the pitch angle of the UAV (the rotation angle around the x-axis); θ y is the roll angle of the UAV (the rotation angle around the y-axis).

[0058] Step S2, perform hierarchical processing on the partial discharge signal through a multi-scale dynamic convolution network, capture features of different frequency bands through a dynamically adjusted convolution kernel group, and perform non-linear suppression on high-frequency noise through an adaptive wavelet threshold function; define the convolution kernel group and use the multi-scale dynamic convolution network to perform hierarchical processing on the original signal, capture features of different frequency bands through a dynamically adjusted convolution kernel group, and combine an improved adaptive wavelet threshold function to perform non-linear suppression on high-frequency noise.

[0059] In one embodiment, the convolution kernel group includes

[0060]

[0061] Among them, W i is the base convolution kernel (a learnable parameter matrix with a size of 3×3 to 7×7); σ is the Sigmoid activation function; α i , β i are learnable weight parameters; s is the input signal feature. The formula dynamically adjusts the contribution weights of different convolution kernels through the Sigmoid gating mechanism.

[0062] The adaptive wavelet threshold function includes

[0063]

[0064] Among them, λ adapt is the dynamically adjusted threshold; N is the signal length; E b is the background noise energy. This threshold is adaptively adjusted according to the noise level.

[0065] The adaptive wavelet threshold function also includes adjusting the wavelet coefficients according to the following rules

[0066]

[0067] where \(w_j\) is the wavelet coefficient; \(\sigma\) n is the noise standard deviation (calculated by the robust estimation method). The advantage is that it hard-thresholds the strong noise coefficients, softens the weak noise coefficients exponentially, and retains the detailed features.

[0068] In one embodiment, the non-linear suppression of high-frequency noise includes performing multi-(set to 7 in this embodiment) layer wavelet packet decomposition on the partial discharge signal, and applying a dynamic convolution kernel group to the multi-(3 - 5) layer high-frequency coefficients, and fusing features through the following cross-scale attention mechanism:

[0069]

[0070] where \(Q\), \(K\), \(V\) are the query, key, and value matrices from different wavelet decomposition layers respectively; \(d\) k is the dimension of the key vector (used to scale the dot product to prevent gradient vanishing). It establishes the correlation between features of different scales and suppresses cross-band interference.

[0071] Step S3, extract the feature vectors in the time domain and frequency domain of the partial discharge signal, and perform discharge type identification according to the fusion result of the time domain and frequency domain feature vectors; construct a time-frequency domain joint feature vector and synchronously extract the Teager energy operator, Hilbert marginal spectrum, and Wigner-Ville distribution features of the signal, and construct a hybrid classification model to achieve discharge type identification.

[0072] In one embodiment, the time domain and frequency domain feature vectors are fused according to the following formula

[0073] \(F = [TKEO(s(t)), HHT(f), Re(WVD(t,f))]\)

[0074] where \(F\) is the joint feature vector (high-dimensional feature matrix), which fuses the time domain transient features, frequency domain energy distribution, and time-frequency resolution, and provides a multi-dimensional data basis for discharge pattern classification; \(TKEO\) is the Teager-Kaiser energy operator (used to capture transient shocks); \(HHT\) is the Hilbert-Huang transform marginal spectrum; \(WVD\) is the real part of the Wigner-Ville distribution; extract features from different angles; \(s(t)\) is the original time domain signal; \(t\) is the time variable; \(f\) is the frequency variable (unit: Hz).

[0075] Specifically, through the random forest - support vector machine hybrid model classification, discharge type identification is performed according to the following formula

[0076]

[0077] where is the predicted discharge type category label (such as corona discharge, surface discharge, internal discharge, etc.); SVM(·) is the support vector machine classifier; RF embedding is the feature embedding function of the random forest; F is the time-frequency domain joint feature vector.

[0078] Step S4: According to the time difference positioning and energy gradient attenuation within the UAV cluster, determine the precise coordinate positioning of the discharge source in three-dimensional space to obtain the final power grid partial discharge detection result. Through the joint optimization of the time difference positioning equation and the energy gradient attenuation model, the precise coordinate calculation of the discharge source is completed in three-dimensional space, forming a full closed-loop processing link from signal acquisition, intelligent denoising, pattern recognition to spatial positioning.

[0079] In one embodiment, the time difference positioning and energy gradient attenuation include performing time difference positioning according to the following formula:

[0080]

[0081] where, Δt ij is the time difference of signal arrival between UAV i and j; r i , r j are the three-dimensional space coordinates (vector form) of UAV i and j; P is the three-dimensional coordinates of the discharge source to be located (partial discharge source); v is the signal propagation speed (depending on the signal type);

[0082] Perform energy gradient attenuation constraint according to the following formula:

[0083]

[0084] where, is the energy gradient (three-dimensional vector); E k is the energy of the discharge signal received by the k-th sensor (UAV); rk is the three-dimensional coordinates (vector form) of the k-th sensor; P is the three-dimensional coordinates (vector form) of the discharge source to be solved.

[0085] The determination of the precise coordinate positioning of the discharge source includes performing positioning by jointly considering time difference positioning and energy gradient attenuation according to the following formula:

[0086]

[0087] where, P is the three-dimensional coordinates (vector form) of the discharge source to be solved; Δt ij is the time difference of signal arrival between UAV i and j; is the square of the L2 norm of the energy gradient (regularization constraint term); λ is the regularization coefficient (weight for balancing time difference positioning and energy constraint); The Levenberg-Marquardt algorithm is used for solution, and the positioning error < 0.5m.

[0088] In summary, implementing the embodiments of the present invention has the following beneficial effects:

[0089] For the method for detecting partial discharge in the power grid provided by the present invention, by constructing an unmanned aerial vehicle (UAV) collaborative sensing network, performing motion compensation preprocessing, and applying a multi-scale dynamic convolution denoising algorithm, dual-domain feature fusion recognition, and time difference-energy gradient joint positioning technology, precise detection of partial discharge carried by the UAV is achieved, the problem of partial discharge detection in a complex electromagnetic environment is solved, and the detection accuracy and positioning accuracy are improved.

[0090] The above-disclosed are only the preferred embodiments of the present invention. Of course, the scope of the rights of the present invention cannot be limited thereby. Therefore, equivalent changes made according to the claims of the present invention still fall within the scope covered by the present invention.

Claims

1. A method for detecting partial discharge in a power grid, characterized in that: include: The local discharge signal is detected by a preset drone cluster, and the collected local discharge signal is corrected by motion compensation phase correction; The local discharge signal is processed in layers through a multi-scale dynamic convolutional network, the characteristics of different frequency bands are captured through a dynamically adjusted convolution kernel group, and the high-frequency noise is nonlinearly suppressed through an adaptive wavelet threshold function. Extract the time domain and frequency domain feature vectors of the partial discharge signal, and identify the discharge type based on the fusion results of the time domain and frequency domain feature vectors; According to the time difference positioning and energy gradient attenuation within the UAV cluster, the precise coordinate positioning of the discharge source is determined in three-dimensional space to obtain the final power grid partial discharge detection results.

2. The method according to claim 1, characterized in that It also includes performing motion compensation phase correction according to the following formula: Among them, S raw (t) is the original sensor signal; Δd is the displacement change of the UAV; λ is the signal wavelength.

3. The method according to claim 2, characterized in that The convolution kernel group includes: Among them, W i is the base convolution kernel; σ is the Sigmoid activation function; α i ,β i is the learnable weight parameter; s is the input signal feature.

4. The method according to claim 3, characterized in that The adaptive wavelet threshold function includes: Among them, λ adapt is the dynamically adjusted threshold; N is the signal length; E b is the background noise energy.

5. The method according to claim 4, characterized in that The adaptive wavelet threshold function also includes adjusting the wavelet coefficients according to the following rules: Among them, wj is the wavelet coefficient; σ n is the noise standard deviation.

6. The method according to claim 5, characterized in that The nonlinear suppression of high-frequency noise includes performing multi-layer wavelet packet decomposition on the local discharge signal, applying a dynamic convolution kernel group to the multi-layer high-frequency coefficients, and fusing features through the following cross-scale attention mechanism: Among them, Q, K, and V are the query, key, and value matrices from different wavelet decomposition layers respectively; d k is the dimension of the key vector.

7. The method according to claim 6, characterized in that It also includes fusing the feature vectors in the time domain and frequency domain according to the following formula: F=[TKEO(s(t)),HHT(f),Re(WVD(t,f))] Among them, F is the joint eigenvector; TKEO is the Teager-Kaiser energy operator; HHT is the Hilbert-Huang transform marginal spectrum; WVD is the real part of Wigner-Ville distribution; s(t) is the original time domain signal; t is the time variable; f is the frequency variable.

8. The method according to claim 7, characterized in that It also includes the identification of discharge type according to the following formula: in, is the predicted discharge type category label; SVM(·) is the support vector machine classifier; RF embedding is the feature embedding function of random forest; F is the joint feature vector in time-frequency domain.

9. The method according to claim 8, characterized in that The time difference positioning and energy gradient attenuation include performing time difference positioning according to the following formula: Among them, Δt ij is the arrival time difference of signals from drones i and j; r i ,r j are the three-dimensional spatial coordinates of UAVs i and j; P is the three-dimensional coordinates of the discharge source to be located; v is the signal propagation speed; The energy gradient decay constraint is performed according to the following formula: in, is the energy gradient; E k is the discharge signal energy received by the kth sensor; r k is the three-dimensional coordinate of the kth sensor; P is the three-dimensional coordinate of the discharge source to be solved.

10. The method according to claim 9, characterized in that Determining the precise coordinates of the discharge source includes combining the time difference positioning and the energy gradient attenuation to perform positioning according to the following formula: Where P is the three-dimensional coordinate of the discharge source to be solved; Δt ij is the arrival time difference of signals from UAV i and j; is the L2 norm square of the energy gradient; λ is the regularization coefficient.

Images