A method and system for detecting defects of distribution network fittings

Abstract

This invention discloses a method and system for detecting defects in power distribution network fittings. The method includes preprocessing and dividing the collected vibration signals of power distribution network fittings into datasets; designing a multi-scale time-frequency domain neural network core structure; constructing a multi-scale time-frequency domain neural network model; and obtaining the output results of the test set samples based on the multi-scale time-frequency domain neural network. The power distribution network fitting defect detection method and system disclosed in this invention provide multi-time-frequency scale analysis of nonlinear and non-stationary fitting vibration signals, which can dynamically capture the characteristics of the signal in the time and frequency domains, thereby improving the detection accuracy of fitting defects and meeting the practical needs for accuracy and speed.

Description

Technical Field

[0001] This invention relates to the field of power distribution network fitting defect detection technology, specifically to a method and system for detecting power distribution network fitting defects. Background Technology

[0002] Connection fittings and protective fittings are critical components in power distribution networks, playing a decisive role in the safe operation of the grid. Connection fittings are not only a fundamental part of the power system but also responsible for power transmission and distribution. Their functionality directly affects the stability and security of the entire power grid. Good connection fittings ensure stable connections between electrical equipment, effectively preventing electrical accidents and line faults. Meanwhile, protective fittings, as protective devices for the power grid, protect the grid and its connected equipment from damage in the event of overloads, short circuits, and ground faults. Therefore, ensuring the reliability and stability of connection and protective fittings is crucial for the smooth operation of the distribution network and requires strict management and continuous optimization. Through in-depth research and improvement of these fittings, we can enhance the reliability and security of the power grid, meeting the ever-growing demand for a stable power supply in modern society.

[0003] Traditional methods for detecting defects in power distribution network fittings include manual visual inspection, infrared and ultraviolet detection, sound detection, and physical and electrical performance measurements. These methods typically require regular execution by professional inspectors and rely on manual operation and visual observation, such as visually inspecting fittings for obvious damage like cracks, corrosion, or deformation. While these methods are simple and direct, they generally lack high accuracy and cannot provide early fault warnings. Furthermore, due to their heavy reliance on manual inspections, these traditional methods are slow and inefficient, and consume a significant amount of manpower, especially in vast or geographically complex power distribution network areas.

[0004] With technological advancements, neural networks, especially deep learning methods, have been applied to defect detection in power distribution network fittings, primarily through image recognition technology. These methods utilize deep learning models such as convolutional neural networks to analyze images captured by surveillance cameras or drones, automatically identifying various defects in the fittings. However, this image-based neural network approach faces several challenges, such as a limited number of training samples and the small size of the fitting targets in the image, resulting in low resolution and affecting detection accuracy. Furthermore, processing large numbers of high-resolution images is slow, which becomes a limiting factor in applications requiring real-time or near-real-time responses.

[0005] The metal fitting image small sample recognition algorithm proposed by Xie Zhihui et al. from the School of Automation and Electronic Engineering of Qingdao University of Science and Technology expands the dataset by generating adversarial networks and introduces foreground enhancement and attention mechanisms to enhance feature extraction. However, this method faces problems such as large differences between generated samples and actual samples, noise introduced by background masks, and increased computational complexity affecting model efficiency and accuracy.

[0006] Zhao Zhenbing from the Department of Electronic and Communication Engineering at North China Electric Power University proposed a method for detecting transmission line fittings and their defects based on contextual structure reasoning. This method combines structural knowledge of transmission line fittings and their defects with bidirectional gated recurrent units and a self-attention mechanism to improve the confidence of correct positive samples and reduce the confidence of incorrect positive samples, thereby improving the overall average accuracy. However, this method faces challenges such as high computational resource consumption, slow processing speed, and the difficulty of adjusting and maintaining complex models.

[0007] In the prior art, an invention patent with publication number CN116992280A discloses a bearing fault diagnosis method and system based on a multi-scale wavelet thresholding network. This method inputs a one-dimensional bearing fault vibration signal from a training set into a fault diagnosis model for training. The training process of the fault diagnosis model includes: frequency division analysis of the bearing fault vibration signal using a multi-scale convolutional wavelet decomposition network; wherein the multi-scale convolutional wavelet decomposition network includes a single-path convolutional wavelet decomposition network with three convolutional kernels of different scales; convolutional wavelet decomposition networks of different paths are used to extract time-frequency features with different characteristics; further feature extraction is performed on the output features of the multi-scale convolutional wavelet decomposition network using a parallel network DRSN-BiGRU consisting of a deep residual shrinking network and BiGRU; and the output features of DRSN-BiGRU are classified using fully connected layers and dropout layers. However, wavelet transform cannot be dynamically adjusted, and the parameters of the wavelet decomposition network are preset values. The combination of convolutional wavelet decomposition networks and BiGRU networks focuses more on utilizing the multi-scale capabilities of wavelet analysis and the bidirectional information flow of BiGRU. It is suitable for applications that require comprehensive analysis of signal history and future trends, but not for local representation. Summary of the Invention

[0008] The technical problem to be solved by the present invention is to provide a rapid and accurate method for detecting defects in hardware.

[0009] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0010] A method for detecting defects in power distribution network fittings includes: using real-time vibration signals of power distribution network fittings as input to an optimal multi-scale time-frequency domain neural network, and the optimal multi-scale time-frequency domain neural network outputting the defect type of the power distribution network fittings;

[0011] The optimal multi-scale time-frequency domain neural network is obtained through the following steps:

[0012] Construct a training set of vibration signals for power distribution network fittings;

[0013] An initial multi-scale time-frequency domain neural network is established, including an adaptive filter bank decomposition layer, a multi-time-frequency scale encoder layer, a frequency domain graph convolutional layer, and a hardware defect feature detection layer;

[0014] The defect x of the distribution network fittings at sampling time t in the training set is given by the k-th item. k,t The data is input into the adaptive filter bank decomposition layer to obtain the k-th distribution network fitting defect data x. k,t Intrinsic information F of distribution network fitting defects at time step t k,t ;

[0015] The intrinsic information of the defects in the power distribution network fittings F k,t Input into the multi-time-frequency scale Encoder layer to obtain the modulation features R of distribution network fitting defects. k,t ;

[0016] Take the k-th data point at time t in the training set as x k,t In the input mid-frequency domain convolutional layer, the spectral features H are obtained. l ;

[0017] Modulation characteristics of distribution network fitting defects R k,t and spectral characteristics H l Input into the hardware defect feature detection layer, for the k-th data at sampling time t, x k,t Defect identification and classification of power distribution network fittings;

[0018] Based on the training set of vibration signals of power distribution network fittings and combined with optimization algorithms, the initial multi-scale time-frequency domain neural network is optimized to obtain the optimal multi-scale time-frequency domain neural network.

[0019] Advantages: This invention utilizes a deep learning network framework and comprehensively considers the defect characteristics of distribution network fittings. An adaptive filter bank decomposition layer is used to initially process the input vibration signal data of the distribution network fittings. Intrinsic information and encoded features of the defects in the distribution network fittings are extracted through a multi-time-frequency scale encoder layer and a frequency domain graph convolutional layer. Next, a multi-head wavelet attention mechanism layer analyzes the dynamic gating information at different scales, and a residual connection feedforward neural network layer enhances the model's learning ability. Finally, a softmax layer outputs the final fault type prediction.

[0020] The model's input is distribution network fitting defect data. After initial processing by an adaptive filter bank decomposition layer, the data enters a multi-time-frequency scale Encoder layer. This layer uses wavelet transforms at different time and frequency scales to extract deeper fault features. Then, these features are fed into a frequency domain graph convolution layer to construct a spectrogram and perform graph convolution operations to obtain more refined spectral features. Next, a multi-head wavelet attention mechanism layer dynamically adjusts and integrates features at different scales. Finally, a residual connection feedforward neural network layer further modulates these features, and a Softmax layer calculates the probability values ​​of all fault types corresponding to the input fault data to obtain the final fault label.

[0021] In one embodiment of the present invention, a training set of vibration signals for power distribution network fittings is constructed and obtained through the following steps:

[0022] Vibration signal data of different power distribution network fittings are collected, and a set of fitting defect classifications is constructed, denoted as X, where X = {X1…X2}. k …X K}, X k The vibration signal represents the k-th hardware defect data, and X k ={x k,1 …x k,t …x k,T}, x k,t The value represents the vibration amplitude at sampling time t for the k-th hardware defect data; 1≤k≤K, where K represents the total number of hardware defects; 1≤t≤T, where T represents the total sampling time.

[0023] Construct a set Y of defect categories for distribution network fittings, denoted as Y = {y1, ..., y2}. k ,…,y K}, where y k Let y represent the label value of the k-th distribution network hardware defect data, and y k It belongs to [1, N], where N is the number of types of hardware defects;

[0024] The labeled power distribution hardware defect dataset P=(X,Y) is randomly shuffled and used as the training set for power distribution hardware vibration signals.

[0025] In one embodiment of the present invention, the adaptive filter bank decomposition layer uses (1)-(5) to perform adaptive filtering processing to obtain the kth distribution network hardware defect data x. k,t Intrinsic information F of distribution network fitting defects at time step t k,t :

[0026] P k,t =|x k,t | 2 #(1)

[0027]

[0028] α k,t =Sigmoid(W p ·P k,t +W ω ·ω k,t +b α )#(4)

[0029]

[0030] In the formula, ω is the frequency, typically ranging from 0 to 2π; M is the length of the window function, where m belongs to [-M, M]; α k,t P is the adjustment function for the k-th distribution network hardware defect data at time t; k,t This is the kth distribution network hardware defect data x k,t Instantaneous energy at sampling time t; Hilbert represents the Hilbert transform; The result of the Hilbert transform is represented by ∠; ω represents the phase angle. k,t W is the instantaneous frequency. p It is P k,t The weight matrix W ω It is ω k,t The weight matrix; b α It is the bias; Sigmoid is a function, e is the natural constant, and j is the imaginary unit.

[0031] In one embodiment of the present invention, the modulation feature R of the distribution network fitting defect is... k,t Obtain it through the following methods:

[0032] The multi-time-frequency scale encoder layer consists of a multi-scale wavelet attention mechanism layer and a residual connection feedforward neural network layer connected in sequence;

[0033] In the multi-scale wavelet attention mechanism layer:

[0034] Extract intrinsic information F of defects in power distribution network fittings k,t Multi-scale time-frequency features are used to obtain the k-th hardware defect data x. k,t Distribution network fitting defect coding features at time step t and scale s.

[0035] Based on the metal fitting defect coding characteristics at the s-th scale Obtain the defect data x of the kth distribution network fittings k,t The dynamic gating G at time step t k,t ;

[0036] By splicing together the hardware defect coding features at all scales, the k-th hardware defect data x is obtained. k,t Defect clustering characteristics of distribution network fittings at time step t k,t ;

[0037] Residual connection feedforward neural network layer:

[0038] Input the agglomeration feature M of the distribution network fitting defects k,t into the residual connection feedforward neural network layer to obtain the k-th fitting defect data x k,t The modulation feature R of the distribution network fitting defects at the t-th time step k,t .

[0039] In an embodiment of the present invention, the k-th fitting defect data x k,t The encoding feature of the fitting defects at the s-th scale at the t-th time step is obtained by using equations (6)-(9):

[0040] γ′ s =γ s ·(1 + η s ·sigmoid(W γ ·F k,t +b γ ))#(6)

[0041] λ′ s =λ s ·(1 + ξ s ·sigmoid(W λ ·F k,t +b λ ))#(7)

[0042]

[0043] In the formula, u is the wavelet time factor, 1 < u ≤ U; s is the wavelet scale factor; x k,t+u is the distribution fitting defect data at the k-th and (t + u)-th moments; ψ′ s (t, u; γ′ s , λ′ s ) is the dynamically adjusted wavelet basis function; ω s is the corresponding scale angular frequency; γ′ s is the amplitude adjustment parameter; λ′ s is the scale adjustment parameter; γ s is the initial amplitude parameter; λ s is the initial scale adjustment parameter; W γ and b γ are the amplitude weight and amplitude bias; η s is the amplitude adjustment factor; ξ s is the scale adjustment factor, Sigmoid is the function, e is the natural constant, j is the imaginary unit, and ψ is the wavelet basis function.

[0044] In an embodiment of the present invention, the k-th distribution network fitting defect data xk,t The dynamic gating G at time step t k,t Using equations (10)-(11), we can obtain:

[0045]

[0046] In the formula, This is the kth distribution network hardware defect data x k,t The signal energy at time step t and scale s, where S is the maximum decomposition scale; log is a logarithmic function; β s and σ s Trainable weights that control the effects of energy and coefficient of variation; yes The variance.

[0047] In one embodiment of the present invention, the kth distribution network fitting defect data x k,t Defect clustering characteristics of distribution network fittings at time step t k,t Using equation (12), we can obtain:

[0048]

[0049] In the formula, Concat is the concatenation function; Attention represents the dot product scaling attention mechanism; Q s ,K s V s The query value, key value, and truth value in the attention mechanism are obtained by linear transformation; S is the maximum decomposition scale.

[0050] In one embodiment of the present invention, the spectral feature H is obtained. l include:

[0051] Based on the defect data of power distribution network fittings x k,t Obtain the spectral amplitude S of the defect in the distribution network fittings. k,f ;

[0052] According to the spectrum amplitude S of the defect in the distribution network fittings k,f Construct a normalized adjacency matrix for the defect spectrum of distribution network fittings.

[0053] Based on the normalized adjacency matrix Obtain spectral features H l .

[0054] In one embodiment of the present invention, the k-th data at sampling time t is... k,t Defect identification and classification of power distribution network fittings are performed using equations (19)-(20), which yield the following results:

[0055] F com =Concat(H L ,Rk,t )#(19)

[0056]

[0057] In the formula, H L F represents the feature matrix of the nodes in the Lth layer; com It is a characteristic of the fusion of defects in power distribution network fittings; W d It is the weight matrix of the power distribution network hardware defect feature detection layer; b d It is the bias term for the predicted distribution network hardware defect type label; Concat is the concatenation function; Softmax is the activation function; Labels for predicted distribution network hardware defect types.

[0058] The present invention also provides a system for detecting defects in power distribution network fittings according to the above-described method, comprising:

[0059] Execution module: Used to take the real-time vibration signal of the distribution network fittings as the input of the optimal multi-scale time-frequency domain neural network, and the optimal multi-scale time-frequency domain neural network outputs the defect type of the distribution network fittings;

[0060] Building blocks: Used to construct optimal multi-scale time-frequency domain neural networks, including the following steps:

[0061] Construct a training set of vibration signals for power distribution network fittings;

[0062] An initial multi-scale time-frequency domain neural network is established, including an adaptive filter bank decomposition layer, a multi-time-frequency scale encoder layer, a frequency domain graph convolutional layer, and a hardware defect feature detection layer;

[0063] The defect x of the distribution network fittings at sampling time t in the training set is given by the k-th item. k,t The data is input into the adaptive filter bank decomposition layer to obtain the k-th distribution network fitting defect data x. k,t Intrinsic information F of distribution network fitting defects at time step t k,t ;

[0064] The intrinsic information of the defects in the power distribution network fittings F k,t Input into the multi-time-frequency scale Encoder layer to obtain the modulation features T of distribution network fitting defects. k,t ;

[0065] Take the k-th data point at time t in the training set as x k,t In the input mid-frequency domain convolutional layer, the spectral features H are obtained. l ;

[0066] Modulation characteristics of distribution network fitting defects R k,t and spectral characteristics H l Input into the hardware defect feature detection layer, for the k-th data at sampling time t, xk,t Defect identification and classification of power distribution network fittings;

[0067] Based on the training set of vibration signals from power distribution network fittings and combined with optimization algorithms, the initial multi-scale time-frequency domain neural network is optimized to obtain the optimal multi-scale time-frequency domain neural network.

[0068] Compared with the prior art, the beneficial effects of the present invention are:

[0069] This invention aims to improve the processing capability of vibration signals from different environments and types of fittings by adaptively adjusting the filter bank and multi-time-frequency scale encoder layer based on the characteristics of the vibration signals from power distribution network fittings. A multi-head wavelet attention mechanism layer is proposed to enhance the multi-scale time-frequency feature modeling capability. Furthermore, a frequency domain graph convolution layer is proposed to construct fitting defect features from the spectrum graph through convolution of the frequency domain feature calculation graph of the fitting vibration signal. Dynamic gating is proposed to further filter the features of complex fitting vibration signals to obtain accurate fitting defect features. Through the precise localization and construction of time-frequency features, the efficiency and accuracy of fitting defect detection are significantly improved.

[0070] An adaptive filter bank and a multi-time-frequency scale encoder layer are used to automatically adjust the filtering and encoding process according to the characteristics of the vibration signal, which improves the ability to handle various field environments and different types of hardware defects. Furthermore, a frequency domain graph convolutional layer further enhances the extraction of frequency domain features and optimizes the identification of defect features in the spectrum graph. Dynamic gating and multi-head wavelet attention mechanisms improve the system's detection speed and accuracy in complex signal environments, significantly improving the efficiency and accuracy of defect detection by precisely locating and emphasizing key time-frequency features.

[0071] It can dynamically capture the characteristics of hardware vibration signals at different time and frequency scales, which makes it possible to analyze and identify defects more accurately for nonlinear and non-stationary vibration signals. This multi-scale analysis provides richer information than traditional single-scale analysis, thereby improving the accuracy and reliability of detection.

[0072] This invention dynamically adjusts the wavelet basis according to the specific characteristics of the signal, making the signal decomposition more adaptable to the current signal state. This method optimizes the local representation of the signal by adjusting wavelet parameters, providing greater flexibility and adaptability. By dynamically adjusting the combination of wavelet transform and attention mechanisms, it focuses more on adaptive signal processing and the extraction of key features, making it suitable for scenarios with large dynamic changes and where specific signal features need to be emphasized.

[0073] This invention focuses on frequency domain characteristics, converting signals to the frequency domain using Fast Fourier Transform and processing spectral data. This method can intuitively capture the periodicity and frequency distribution characteristics of signals, making it particularly suitable for analyzing highly periodic data such as vibrations, sound, or electrical signals. Attached Figure Description

[0074] Figure 1 This is a schematic diagram of a method for detecting defects in power distribution network fittings according to an embodiment of the present invention.

[0075] Figure 2 This is a block diagram of a power distribution network hardware defect detection system according to an embodiment of the present invention. Detailed Implementation

[0076] To facilitate understanding of the technical solution of the present invention by those skilled in the art, the technical solution of the present invention will now be further described in conjunction with the accompanying drawings.

[0077] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0078] Please see Figure 1 As shown, the present invention provides a method for detecting defects in distribution network fittings, including using real-time vibration signals of distribution network fittings as input to an optimal multi-scale time-frequency domain neural network, and the optimal multi-scale time-frequency domain neural network outputting the defect type of the distribution network fittings.

[0079] In one embodiment of the present invention, the optimal multi-scale time-frequency domain neural network is obtained through the following steps:

[0080] Construct a training set of vibration signals for power distribution network fittings.

[0081] An initial multi-scale time-frequency domain neural network is established, including an adaptive filter bank decomposition layer, a multi-time-frequency scale encoder layer, a frequency domain graph convolutional layer, and a hardware defect feature detection layer.

[0082] The defect x of the distribution network fittings at sampling time t in the training set is given by the k-th item. k,t The data is input into the adaptive filter bank decomposition layer to obtain the k-th distribution network fitting defect data x. k,t Intrinsic information F of distribution network fitting defects at time step t k,t .

[0083] The intrinsic information of the defects in the power distribution network fittings F k,t Input into the multi-time-frequency scale Encoder layer to obtain the modulation features R of distribution network fitting defects.k,t .

[0084] Take the k-th data point at time t in the training set as x k,t In the input mid-frequency domain convolutional layer, the spectral features H are obtained. l .

[0085] Modulation characteristics of distribution network fitting defects R k,t and spectral characteristics H l Input into the hardware defect feature detection layer, for the k-th data at sampling time t, x k,t Defect identification and classification are performed on the power distribution network fittings.

[0086] Based on the training set of vibration signals of power distribution network fittings and combined with optimization algorithms, the initial multi-scale time-frequency domain neural network is optimized to obtain the optimal multi-scale time-frequency domain neural network.

[0087] In this embodiment, constructing the training set of vibration signals for distribution network fittings includes:

[0088] Vibration signal data of different power distribution network fittings are collected. After normalizing the vibration signals of the power distribution network fittings collected by the installed vibration signal sensors, a set of defect classifications for power distribution network fittings is constructed, denoted as X, where X = {X1…X2}. k …X K}, X k X represents the vibration signal of the defect data of the kth distribution network fitting, and X k ={x k,1 …x k,t …x k,T}, x k,t Let represent the vibration amplitude at sampling time t for the k-th hardware defect data; 1≤k≤K, where K represents the total number of hardware defects; 1≤t≤T, where T represents the total sampling time.

[0089] Construct a set Y of distribution network fitting defect categories, and assign labels to the vibration signals corresponding to the defect types of distribution network fittings based on prior knowledge, denoted as Y = {y1, ..., y2}. k ,…,y K}, where y k Let y represent the label value of the k-th distribution network hardware defect data, and y k It belongs to [1, N], where N is the number of types of defects in power distribution network fittings.

[0090] The labeled power distribution hardware defect dataset P=(X,Y) is randomly shuffled and used as the training set.

[0091] In one embodiment of the present invention, in the adaptive filter bank decomposition layer:

[0092] Data on defective distribution network fittings (item k) k,t Data at sampling time t xk,t The input adaptive filter bank decomposition layer functions to perform frequency decomposition on the vibration signal data of the distribution network hardware, dynamically adjust the filtering parameters to optimize signal processing, and provide accurate basic data for subsequent defect feature extraction and analysis. Adaptive filtering is performed using equations (1)-(5) to obtain the kth hardware defect data x. k,t The intrinsic information F of the hardware defect at time step t. k,t :

[0093] P k,t =|x k,t | 2 #(1)

[0094]

[0095] α k,t =Sigmoid(W p ·P k,t +W ω ·ω k,t +b α )#(4)

[0096]

[0097] In the formula, ω is the frequency, typically ranging from 0 to 2π; M is the length of the window function, where m belongs to [-M, M]; α k,t P is the adjustment function for the k-th distribution network hardware defect data at time t; k,t This is the kth distribution network hardware defect data x k,t Instantaneous energy at sampling time t; Hilbert represents the Hilbert transform; The result of the Hilbert transform is represented by ∠; ω represents the phase angle. k,t W is the instantaneous frequency. p It is P k,t The weight matrix W ε It is ω k,t The weight matrix; b α It is the bias; Sigmoid is a function, e is the natural constant, and j is the imaginary unit.

[0098] In one embodiment of the present invention, the multi-time-frequency scale Encoder layer includes a multi-scale wavelet attention mechanism layer and a residual connection feedforward neural network layer connected in sequence.

[0099] In this embodiment, the multi-scale wavelet attention mechanism layer effectively extracts and enhances the multi-scale time-frequency features of the hardware defect data, thereby enhancing the model's ability to identify key defect features. The k-th hardware defect data x is obtained using equations (6)-(9). k,t Hardware defect coding features at time step t and scale s.

[0100] γ′ s =γ s ·(1 + η s ·sigmoid(W γ ·F k,t +b γ ))#(6)

[0101] λ′ s =λ s ·(1 + ξ s ·sigmoid(W λ ·F k,t +b λ ))#(7)

[0102]

[0103] In the formula, u is the wavelet time factor, 1 < u ≤ U; s is the wavelet scale factor; x k,t+u is the defect data of the distribution fitting hardware at the (t + u)-th moment of the k-th item; ψ′ s (t, u; γ′ s , λ′ s ) is the dynamically adjusted wavelet basis function; ω s is the corresponding scale angular frequency; γ′ s is the amplitude adjustment parameter; λ′ s is the scale adjustment parameter; γ s is the initial amplitude parameter; λ s is the initial scale adjustment parameter; W γ and b γ are the amplitude weight and amplitude bias; η s is the amplitude adjustment factor; ξ s is the scale adjustment factor, Sigmoid is a function, e is the natural constant, j is the imaginary unit, and ψ is the wavelet basis function.

[0104] The multi-head wavelet attention mechanism layer uses equations (10)-(11) to obtain the defect data x k,t of the k-th fitting hardware at the t-th time step of the dynamic gating G k,t :

[0105]

[0106] In the formula, is the signal energy of the defect data x k,t of the k-th distribution fitting hardware at the t-th time step and the s-th scale, S is the maximum decomposition scale; log is the logarithmic function; β s and σ s are the trainable weights that control the influence of energy and coefficient of variation; is The variance. In this embodiment, the multi-head wavelet attention mechanism layer can be understood as multiple wavelet attention mechanism layers, each corresponding to a wavelet scale.

[0107] In this embodiment, the multi-head wavelet attention mechanism layer uses equation (12) to obtain the k-th distribution network fitting defect data x. k,t Defect clustering characteristics of distribution network fittings at time step t k,t :

[0108]

[0109] In the formula, Concat is the concatenation function; Attention represents the dot product scaling attention mechanism; Q s ,K s V s These are the query value, key value, and true value, respectively, obtained through linear transformation; S is the maximum decomposition scale.

[0110] In this embodiment, the residual connection feedforward neural network layer helps the model maintain stable training as its depth increases, improving the efficiency and accuracy of feature processing. The kth distribution network fitting defect data x is obtained using equation (13). k,t Modulation characteristics of distribution network fitting defects at time step t R k,t :

[0111] R k,t =M k,t +W2·ReLU(W1·M k,t +b1)+b2#(13)

[0112] In the formula, W1 is the first-layer modulation weight; b1 is the first-layer modulation bias; W2 is the second-layer modulation weight; b2 is the second-layer modulation bias; and ReLU is the nonlinear activation function.

[0113] Please see Figure 1 As shown, in one embodiment of the present invention, the frequency domain graph convolutional layer includes: hardware defect spectrum calculation, spectrum graph construction, and spectrum graph convolution.

[0114] In this embodiment, the spectrum calculation of the hardware defect reveals the frequency components and characteristics of the signal, providing key frequency domain information for subsequent defect analysis. Specifically, the amplitude S of the hardware defect spectrum is obtained using equation (14). k,f :

[0115] S k,f =FFT(x) k,t )#(14)

[0116] In equation (14), FFT represents Fast Fourier Transform.

[0117] In this embodiment, the spectrogram construction transforms the spectral data of hardware defects into a graph structure and captures and learns the nonlinearity and nonlocal dependencies in the hardware defect data. Specifically, the normalized adjacency matrix of the hardware defect spectrum is constructed using equations (15)-(17).

[0118]

[0119] In the formula, ∥·∥ 2 σ represents the matrix 2-norm; σ represents the mean; D ii S is the degree matrix; k,i and S k,j Let i and j represent the spectral amplitudes of the kth distribution network hardware defect data at frequencies i and j, respectively, and exp is an exponential function.

[0120] In this embodiment, spectrogram convolution aggregates and extracts local and global information from the spectral data, and comprehensively considers the structural features and frequency correlations in the spectrogram. Specifically, spectral features H are extracted using equation (18). l :

[0121]

[0122] In the formula, H l This is the node feature matrix of the l-th layer, i.e., the spectral features, initially S. k,f W l and b l These are the graph convolution weights and biases of the l-th layer, respectively; l is the number of graph convolution layers, 1. <l≤L。

[0123] In one embodiment of the present invention, the power distribution network hardware defect feature detection layer integrates the features extracted from the preceding network layers and uses a classifier to accurately identify and classify hardware defects, thereby enhancing the diagnostic efficacy and accuracy of the model. The hardware defect type label of the kth hardware defect data is obtained using equations (19)-(20).

[0124] F com =Concat(H L ,R k,t )#(19)

[0125]

[0126] In the formula, H L F represents the feature matrix of the nodes in the Lth layer; com It is a characteristic of the fusion of defects in power distribution network fittings; W d It is the weight matrix of the power distribution network hardware defect feature detection layer; b d It is the bias term for the predicted distribution network hardware defect type label; Concat is the concatenation function; Softmax is the activation function; Labels for predicted distribution network hardware defect types.

[0127] In one embodiment of the present invention, a multi-scale time-frequency domain neural network is trained based on a training set of vibration signals of distribution network fittings and using backpropagation or gradient descent. When the number of training rounds reaches the maximum or the loss function reaches the minimum, the training is stopped, thereby obtaining the best trained multi-scale time-frequency domain neural network, which is used to map the input set of distribution network fitting defect data to the corresponding distribution network fitting defect category label.

[0128] Please see Figure 2 As shown, the present invention also provides a system for detecting defects in power distribution network fittings, comprising:

[0129] Execution module: Used to take the real-time vibration signal of the distribution network fittings as the input of the optimal multi-scale time-frequency domain neural network, and output the defect type of the distribution network fittings.

[0130] Building blocks: Used to obtain the optimal multi-scale time-frequency domain neural network.

[0131] Obtaining the optimal multi-scale time-frequency domain neural network includes the following steps:

[0132] Construct a training set of vibration signals for power distribution network fittings;

[0133] An initial multi-scale time-frequency domain neural network is established, including an adaptive filter bank decomposition layer, a multi-time-frequency scale encoder layer, a frequency domain graph convolutional layer, and a hardware defect feature detection layer;

[0134] The defect x of the distribution network fittings at sampling time t in the training set is given by the k-th item. k,t The data is input into the adaptive filter bank decomposition layer to obtain the k-th distribution network fitting defect data x. k,t Intrinsic information F of distribution network fitting defects at time step t k,t ;

[0135] The intrinsic information of the defects in the power distribution network fittings F k,t Input into the multi-time-frequency scale Encoder layer to obtain the modulation features T of distribution network fitting defects. k,t ;

[0136] Take the k-th data point at time t in the training set as x k,t In the input mid-frequency domain convolutional layer, the spectral features H are obtained. l ;

[0137] Modulation characteristics of distribution network fitting defects R k,t and spectral characteristics H l Input into the hardware defect feature detection layer, for the k-th data at sampling time t, x k,tDefect identification and classification of power distribution network fittings;

[0138] Based on the training set of vibration signals from power distribution network fittings and combined with optimization algorithms, the initial multi-scale time-frequency domain neural network is optimized to obtain the optimal multi-scale time-frequency domain neural network.

[0139] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention, and no reference numerals in the claims should be construed as limiting the scope of the claims.

[0140] The above embodiments are merely examples of implementation methods of the invention. The scope of protection of the present invention is not limited to the above embodiments. For those skilled in the art, several modifications and improvements can be made without departing from the concept of the present invention, and these all fall within the scope of protection of the present invention.

Claims

1. A method for detecting defects in power distribution network fittings, characterized in that, include: The real-time vibration signal of the distribution network fittings is used as the input of the optimal multi-scale time-frequency domain neural network, and the optimal multi-scale time-frequency domain neural network outputs the defect type of the distribution network fittings. The optimal multi-scale time-frequency domain neural network is obtained through the following steps: Construct a training set of vibration signals for power distribution network fittings; An initial multi-scale time-frequency domain neural network is established, including an adaptive filter bank decomposition layer, a multi-time-frequency scale encoder layer, a frequency domain graph convolutional layer, and a hardware defect feature detection layer; The adaptive filter bank decomposition layer uses (1)-(5) for adaptive filtering to obtain the first... Data on defects in distribution network fittings In the first Intrinsic information on defects in distribution network fittings at time steps : In the formula, It is the frequency, and its value usually ranges from 0 to 2π; It is the length of the window function. belong[- , ]; It is the first Data on defects in power distribution network fittings Adjust the function at any time; It is the first Data on defects in distribution network fittings No. Instantaneous energy at the sampling moment; Represents the Hilbert transform; Indicate the result of the Hilbert transform; Indicates the phase angle; Instantaneous frequency; yes The weight matrix; yes The weight matrix; It is a bias; For function, Let j be the natural constant and j be the imaginary unit. The training set Article 1 Sampling time distribution network hardware defects The input is fed into the adaptive filter bank decomposition layer to obtain the first... Data on defects in distribution network fittings In the first Intrinsic information on defects in distribution network fittings at time steps ; Intrinsic information on defects in power distribution network fittings Input into the multi-time-frequency scale Encoder layer to obtain the modulation features of distribution network fitting defects. : The multi-time-frequency scale encoder layer consists of a multi-scale wavelet attention mechanism layer and a residual connection feedforward neural network layer connected in sequence; In the multi-scale wavelet attention mechanism layer: Extract intrinsic information of defects in power distribution network fittings Multi-scale time-frequency features, to obtain the first Defect data of metal fittings No. The first step of time Defect coding characteristics of distribution network fittings at scale : According to the Scale of hardware defect coding features , obtain the Data on defects in power distribution network fittings No. Dynamic gating of time steps ; By splicing together the hardware defect coding features at all scales, the first... Defect data of metal fittings No. Time-step characteristics of distribution network hardware defect clustering ; Residual connection feedforward neural network layer: Characteristics of defect clustering in power distribution network fittings The input is fed into the residual connection feedforward neural network layer to obtain the first... Defect data of metal fittings No. Modulation characteristics of distribution network fitting defects at time steps ; The training set Article 1 Sampling time data Spectral features are obtained in the convolutional layer of the input mid-frequency domain image. : Based on the defect data of power distribution network fittings Obtain the spectral amplitude of defects in power distribution network fittings. ; Based on the spectrum amplitude of defects in power distribution network fittings Construct a normalized adjacency matrix for the defect spectrum of distribution network fittings. ; Based on the normalized adjacency matrix Obtain spectral features ; Modulation characteristics of distribution network fitting defects and spectral characteristics Input into the hardware defect feature detection layer, for the first Article 1 Sampling time data Defect identification and classification of power distribution network fittings are performed using equations (19)-(20), which yield the following results: In the formula, Indicates the first Layer node feature matrix; It is a characteristic of the fusion of defects in power distribution network fittings; It is the weight matrix of the defect feature detection layer for power distribution network fittings; It is the bias item of the predicted distribution network hardware defect type label; This is a concatenation function; For activation functions; Labels for predicted distribution network hardware defect types; Based on the training set of vibration signals of power distribution network fittings and combined with optimization algorithms, the initial multi-scale time-frequency domain neural network is optimized to obtain the optimal multi-scale time-frequency domain neural network.

2. The method for detecting defects in power distribution network fittings according to claim 1, characterized in that, The training set of vibration signals for distribution network fittings is constructed through the following steps: Vibration signal data of different power distribution network fittings were collected, and a set of fitting defect classifications was constructed, denoted as . ,and , Indicates the first Vibration signals from metal fitting defect data, and , Indicates the first Defect data of metal fittings The vibration amplitude at the sampling time; K represents the total number of hardware defects; T represents the total sampling time; Construct a set of defect categories for power distribution network fittings , recorded as ,in, Indicates the first The label value of the defect data of the strip distribution network hardware, and It belongs to [1, N], where N is the number of types of hardware defects; Data set of tagged distribution network fitting defects The randomized order was used as a training set for the vibration signals of the distribution network fittings.

3. The method for detecting defects in power distribution network fittings according to claim 1, characterized in that, No. Defect data of metal fittings No. The first step of time Scale of hardware defect coding features Using equations (6)-(9), we can obtain: In the formula, It is a wavelet time factor. ; It is the wavelet scaling factor; It is the first Article 1 Real-time defect data for power distribution hardware; It is a dynamically adjusted wavelet basis function; Corresponding scale angular frequency; It is the amplitude adjustment parameter; It is a scale adjustment parameter; It is the initial amplitude parameter; These are the initial scale adjustment parameters; and These are amplitude weighting and amplitude bias; It is the amplitude adjustment factor; It is a scale adjustment factor. For function, Let j be the natural constant and j be the imaginary unit. These are wavelet basis functions.

4. The method for detecting defects in power distribution network fittings according to claim 1, characterized in that, No. Data on defects in power distribution network fittings No. Dynamic gating of time steps Using equations (10)-(11), we can obtain: In the formula, It is the first Data on defects in distribution network fittings No. Time step Signal energy at the scale The maximum decomposition scale; It is a logarithmic function; and Trainable weights that control the effects of energy and coefficient of variation; yes The variance.

5. The method for detecting defects in power distribution network fittings according to claim 1, characterized in that, No. Data on defects in distribution network fittings No. Time-step characteristics of distribution network hardware defect clustering Using equation (12), we can obtain: In the formula, This is a concatenation function; This represents the dot product scaling attention mechanism; The query value, key value, and truth value in the attention mechanism are obtained by linear transformation. This is the maximum decomposition scale.

6. A system for detecting defects in power distribution network fittings according to any one of claims 1-5, characterized in that, include: Execution module: Used to take the real-time vibration signal of the distribution network fittings as the input of the optimal multi-scale time-frequency domain neural network, and the optimal multi-scale time-frequency domain neural network outputs the defect type of the distribution network fittings; Building blocks: Used to construct optimal multi-scale time-frequency domain neural networks, including the following steps: Construct a training set of vibration signals for power distribution network fittings; An initial multi-scale time-frequency domain neural network is established, including an adaptive filter bank decomposition layer, a multi-time-frequency scale encoder layer, a frequency domain graph convolutional layer, and a hardware defect feature detection layer; The training set Article 1 Sampling time distribution network hardware defects The input is fed into the adaptive filter bank decomposition layer to obtain the first... Data on defects in distribution network fittings In the first Intrinsic information on defects in distribution network fittings at time steps ; Intrinsic information on defects in power distribution network fittings Input into the multi-time-frequency scale Encoder layer to obtain the modulation features of distribution network fitting defects. ; The training set Article 1 Sampling time data Spectral features are obtained in the convolutional layer of the input mid-frequency domain image. ; Modulation characteristics of distribution network fitting defects and spectral characteristics Input into the hardware defect feature detection layer, for the first Article 1 Sampling time data Defect identification and classification of power distribution network fittings; Based on the training set of vibration signals from power distribution network fittings and combined with optimization algorithms, the initial multi-scale time-frequency domain neural network is optimized to obtain the optimal multi-scale time-frequency domain neural network.

Images