CN-121980500-A - Low-voltage power supply cable fault detection method and system

CN121980500ACN 121980500 ACN121980500 ACN 121980500ACN-121980500-A

Abstract

The invention discloses a low-voltage power supply cable fault detection method and a system. The method comprises the steps of obtaining cable fault data, preprocessing the cable fault data, inputting the preprocessed data into a cable fault detection model for classifying and diagnosing the cable faults to obtain detection results, wherein the cable fault detection model is used for acquiring multi-source cable fault data through injecting characteristic frequency signals into a neutral line of a low-voltage power supply cable, preprocessing the multi-source cable fault data to obtain preprocessed signals, carrying out deep characteristic extraction and weighted fusion on the preprocessed signals by utilizing a multichannel one-dimensional convolutional neural network and combining an attention mechanism, realizing fault type identification through a fully-connected network, and outputting the detection results. By implementing the method, the cable fault diagnosis is realized by adopting the deep learning model, so that the accuracy of fault identification is improved, and the excellent performance is shown in the aspects of complexity and nonlinearity.

Inventors

LI CHENYING
CAO JINGXING
ZHANG WEI
ZHOU LI
TAN XIAO
WANG QI
WU SHUQUN
ZHANG YIMING

Assignees

国网江苏省电力有限公司电力科学研究院
国网江苏省电力有限公司
江苏省电力试验研究院有限公司
东南大学溧阳研究院

Dates

Publication Date: 20260505
Application Date: 20260120

Claims (10)

1. The low-voltage power supply cable fault detection method is characterized by comprising the following steps of: Acquiring cable fault data; preprocessing the cable fault data to obtain preprocessed data; inputting the preprocessed data into a cable fault detection model to conduct classification diagnosis of the cable faults so as to obtain detection results, wherein the cable fault detection model is used for injecting characteristic frequency signals into a neutral line of a low-voltage power supply cable, collecting multi-source cable fault data through a coupling transformer, preprocessing the cable fault data to obtain preprocessed signals, conducting deep feature extraction and weighted fusion on the preprocessed signals by utilizing a multichannel one-dimensional convolutional neural network and combining an attention mechanism, and realizing fault type identification through a fully-connected network; And outputting the detection result.
2. The method for detecting a fault in a low-voltage power supply cable according to claim 1, wherein the step of inputting the preprocessed data into a cable fault detection model to perform classification diagnosis of the fault in the low-voltage power supply cable to obtain a detection result comprises: Inputting the preprocessed data into a cable fault detection model, and respectively extracting deep features of different modes in the preprocessed data to obtain multi-channel features; weighting and fusing the multichannel characteristics by adopting an attention mechanism to obtain fused characteristics; and inputting the fusion characteristics into a fully-connected network to perform fault type identification so as to obtain a detection result.
3. The low voltage power cable fault detection method according to claim 1, wherein the training process of the cable fault detection model comprises: Injecting an alternating current voltage signal with characteristic frequency into a distribution network system through a primary side of a bus transformer, injecting a signal with frequency different from power frequency, coupling the injected signal to a secondary side of the transformer, and entering the distribution network system through a neutral point of a three-phase power supply to form a zero-sequence voltage signal; According to the distribution rule of the zero sequence voltage signals in the circuit, on-line calculation is carried out on the cable grounding faults and the ground distribution capacitance, fault current signals are collected on the circuit from the neutral point of the transformer to the grounding, three-phase voltage data of a cable outlet are collected, and the ground impedance is calculated according to kirchhoff's law, so that cable fault data are obtained.
4. The method for detecting a fault in a low-voltage power supply cable according to claim 3, wherein after performing online calculation on a cable ground fault and a distributed capacitance to ground according to a distribution rule of the zero-sequence voltage signals in a line, collecting fault current signals on a line from a neutral point of a transformer to a ground, collecting three-phase voltage data of a cable outlet, and calculating impedance to ground according to kirchhoff's law to obtain multi-source cable fault data, the method further comprises: and carrying out signal denoising, signal normalization and signal expansion on the multi-source cable fault data to obtain a preprocessed signal.
5. The method for detecting a fault in a low voltage power supply cable according to claim 4, wherein the performing signal denoising, signal normalization and signal expansion on the multi-source cable fault data to obtain a preprocessed signal comprises: And performing expansion processing on the multi-source cable fault data by adopting an overlap sampling method, arranging the multi-source cable fault data and the corresponding A, B, C three-phase voltage data in the same row, performing normalization processing to construct a one-dimensional data set, and forming a multi-source fault information database to obtain the preprocessed signals.
6. The method for detecting a fault in a low voltage power supply cable according to claim 5, wherein the denoising process is performed on the zero sequence current signal and the three-phase voltage data, the expansion process is performed on the multi-source cable fault data by adopting an overlap sampling method, the multi-source cable fault data and the corresponding A, B, C three-phase voltage data are arranged in the same row, the normalization process is performed to construct a one-dimensional data set, and a multi-source fault information database is formed to obtain a preprocessed signal, and then the method further comprises: The preprocessed signals share weight in a local area through convolution operation, input layer data are formed by connecting original fault current and voltage data after random overlapping sampling, and the input layer data are equally divided into four samples, and the four samples respectively represent fault current and A, B, C three-phase voltage information; Inputting the four samples into four parallel channels respectively for multi-channel feature extraction, wherein the parameters of the convolution layers of each channel are set identically, and the four parallel channels comprise a plurality of convolution layers and a pooling layer; inputting the extracted multi-channel characteristics into a full-connection layer, and capturing differences among various categories to obtain signal characteristics; and classifying the signal characteristics through a Softmax classification layer, and outputting the classification probability of each fault class.
7. The method for detecting a fault in a low voltage power supply cable according to claim 6, wherein the inputting the four samples into four parallel channels for multi-channel feature extraction includes: each sample is converted into a scalar by global averaging pooling and combined to form a one-dimensional scalar vector, the one-dimensional scalar vector is converted into a one-dimensional weight vector through excitation operation, the one-dimensional weight vector is multiplied with an input sample element by element through rescaling operation, and the normalized weight is applied to each channel.
8. The low voltage power cable fault detection method according to claim 7, wherein the converting each sample from global average pooling to scalar comprises: By using The samples are converted to scalar quantities, wherein, Is that Data of the channel; Is a scalar converted by the channel attention mechanism.
9. The method of claim 7, wherein the converting the one-dimensional scalar vector into a one-dimensional weight vector via the excitation operation comprises: By using A one-dimensional weight vector is determined, wherein, Is that A function; Is that A function; , Is a parameter of the full connection layer; , is a bias term that is used to determine, Is a one-dimensional scalar vector; is a one-dimensional weight vector.
10. The utility model provides a low voltage power supply cable fault detection system which characterized in that includes: the acquisition unit is used for acquiring cable fault data; The preprocessing unit is used for preprocessing the cable fault data to obtain preprocessed data; the diagnosis unit is used for inputting the preprocessed data into the cable fault detection model to carry out classified diagnosis on the cable faults so as to obtain detection results; the output unit is used for outputting the detection result; The cable fault detection model is characterized in that characteristic frequency signals are injected into a neutral line of a low-voltage power supply cable, multi-source cable fault data are collected through a coupling transformer, the signals are preprocessed to obtain preprocessed signals, deep feature extraction and weighted fusion are carried out on the preprocessed signals by utilizing a multichannel one-dimensional convolutional neural network and combining an attention mechanism, and fault type identification is achieved through a fully-connected network.

Description

Low-voltage power supply cable fault detection method and system Technical Field The invention relates to the technical field of fault diagnosis of low-voltage cables, in particular to a fault detection method and system for a low-voltage power supply cable. Background Low voltage power supply cables are a critical component of urban distribution networks and are exposed to complex physical and electrical environments for long periods of time, which makes them susceptible to potential insulation defects. In recent years, with the progress of signal processing technology, particularly the breakthrough of the field of artificial intelligence, the field of cable fault diagnosis has been innovated. By applying advanced signal processing methods such as denoising, modal analysis, feature extraction and deep analysis of signal time-frequency domains, highly relevant feature information can be extracted from fault signals, and powerful technical support is provided for accurate identification of cable faults. Along with the continuous evolution of the smart grid system, a large number of smart devices and sensors are integrated in the power system, so that the level of intellectualization of the system is improved, and the low-voltage line structure in the system for the substation is further complicated. This complexity further enhances the nonlinearity and complexity of the fault signal data, thereby increasing the difficulty in fault identification of the low voltage power supply cable. In addition, conventional methods often rely on extensive manual intervention for data preprocessing and result analysis, which is time and resource consuming and may result in inaccurate diagnostic results due to human factors. Especially in the face of increasing data volume and complexity in modern smart grid systems, the traditional method is worry, and cannot meet the requirements of efficient and accurate fault diagnosis. Therefore, it is necessary to design a new method to realize cable fault diagnosis by adopting a deep learning model, so that not only the accuracy of fault identification is improved, but also superior performance is presented in terms of complexity and nonlinearity problem treatment. Disclosure of Invention The invention aims to overcome the defects of the prior art and provide a low-voltage power supply cable fault detection method and system. In order to achieve the purpose, the invention adopts the following technical scheme that the low-voltage power supply cable fault detection method comprises the following steps: Acquiring cable fault data; preprocessing the cable fault data to obtain preprocessed data; inputting the preprocessed data into a cable fault detection model to conduct classification diagnosis of the cable faults so as to obtain detection results, wherein the cable fault detection model is used for injecting characteristic frequency signals into a neutral line of a low-voltage power supply cable, collecting multi-source cable fault data through a coupling transformer, preprocessing the cable fault data to obtain preprocessed signals, conducting deep feature extraction and weighted fusion on the preprocessed signals by utilizing a multichannel one-dimensional convolutional neural network and combining an attention mechanism, and realizing fault type identification through a fully-connected network; And outputting the detection result. The further technical scheme is that the step of inputting the preprocessed data into a cable fault detection model for classification diagnosis of the cable faults to obtain detection results comprises the following steps: Inputting the preprocessed data into a cable fault detection model, and respectively extracting deep features of different modes in the preprocessed data to obtain multi-channel features; weighting and fusing the multichannel characteristics by adopting an attention mechanism to obtain fused characteristics; and inputting the fusion characteristics into a fully-connected network to perform fault type identification so as to obtain a detection result. The cable fault detection model training process comprises the following steps: Injecting an alternating current voltage signal with characteristic frequency into a distribution network system through a primary side of a bus transformer, injecting a signal with frequency different from power frequency, coupling the injected signal to a secondary side of the transformer, and entering the distribution network system through a neutral point of a three-phase power supply to form a zero-sequence voltage signal; According to the distribution rule of the zero sequence voltage signals in the circuit, on-line calculation is carried out on the cable grounding faults and the ground distribution capacitance, fault current signals are collected on the circuit from the neutral point of the transformer to the grounding, three-phase voltage data of a cable outlet are collected, and the ground impedance is