CN-121995168-A - Partial discharge mode identification method based on ultrahigh frequency signal phase irrelevant features

CN121995168ACN 121995168 ACN121995168 ACN 121995168ACN-121995168-A

Abstract

The invention discloses a partial discharge pattern recognition method based on ultrahigh frequency signal phase irrelevant features, which comprises the steps of generating a phase resolution partial discharge map, obtaining a preprocessing phase resolution partial discharge map with uniform size, forming a phase irrelevant composite feature vector, obtaining a low redundancy phase irrelevant distinguishing feature vector, establishing a corresponding distinguishing energy potential function set after distinguishing a diffusion matrix is stabilized, forming a distinguishing energy potential function model, determining a partial discharge pattern to which a partial discharge sample to be detected belongs according to a minimum distinguishing energy criterion, and outputting a partial discharge pattern recognition result. The self-adaptive dimension reduction method can obviously reduce the interference of noise features and weak discrimination features on the model, and improve the stability and robustness of the discrimination model under the conditions of small samples and short-time monitoring.

Inventors

SUN LEI
REN MING
XIAO HANYAN
LI YUJIE
JIANG ZHIYUAN
YIN ZE
ZHAO KE
WANG LIJIANG
She Congdong
Zhuang Tianxin
LIU JIANJUN
HU CHENGBO
LU YONGLING

Assignees

国网江苏省电力有限公司电力科学研究院
西安交通大学

Dates

Publication Date: 20260508
Application Date: 20260121

Claims (8)

1. The partial discharge mode identification method based on the ultrahigh frequency signal phase independence characteristic is characterized by comprising the following steps of: Collecting an ultrahigh frequency time domain pulse signal by using a UHF receiving device, mapping the ultrahigh frequency time domain pulse signal into amplitude-power frequency phase two-dimensional coordinate data, and generating a phase resolution partial discharge map; Preprocessing the phase resolution partial discharge spectrum, and resampling according to the uniform resolution to obtain a preprocessed phase resolution partial discharge spectrum with uniform size; Extracting an amplitude-phase statistical feature vector, a pulse number-amplitude statistical feature vector, an energy-phase statistical feature vector and a frequency domain related feature vector based on the preprocessing phase resolution partial discharge spectrum, and splicing according to a preset sequence to form a phase irrelevant composite feature vector; Performing self-adaptive feature dimension reduction processing based on category separability contribution degree on the phase irrelevant composite feature vector by using a training data set containing a plurality of marked partial discharge samples to obtain a low redundancy phase irrelevant discriminating feature vector; respectively calculating the discrimination prototype centers and the discrimination diffusion matrixes of different partial discharge modes according to the low-redundancy phase irrelevant discrimination feature vectors, and establishing a corresponding discrimination energy potential function set after the discrimination diffusion matrixes are stabilized to form a discrimination energy potential function model; and inputting the low redundancy phase irrelevant discrimination feature vector of the partial discharge sample to be detected into a discrimination energy potential function model, determining the partial discharge mode of the partial discharge sample to be detected according to the minimum discrimination energy criterion, and outputting a partial discharge mode identification result.
2. The partial discharge pattern recognition method based on the phase independence characteristic of the ultrahigh frequency signal according to claim 1, wherein the mapping of the ultrahigh frequency time domain pulse signal into amplitude-power frequency phase two-dimensional coordinate data comprises the following steps: The UHF receiving device adopting a continuous sampling mode continuously collects electromagnetic radiation signals generated by partial discharge in the gas insulated metal-enclosed switchgear within a preset collection time period to obtain ultra-high frequency time domain pulse signals; Performing equal interval discrete sampling on the ultrahigh frequency time domain pulse signal at a set sampling frequency to obtain a group of ultrahigh frequency discrete sampling sequences; detecting all discharge pulse events in the ultrahigh frequency discrete sampling sequence to form a discharge pulse event set; synchronously acquiring a power frequency reference signal aligned with the ultra-high frequency time domain pulse signal on a time axis, and calculating a power frequency phase corresponding to each discharge pulse event according to the arrival time of each discharge pulse event and the phase information of the power frequency reference signal; mapping all discharge pulse events according to the pulse amplitude and the power frequency phase to obtain a group of two-dimensional coordinates of the amplitude and the power frequency phase, wherein each pair of two-dimensional coordinates of the amplitude and the power frequency phase respectively correspond to the amplitude of one discharge pulse event and the corresponding power frequency phase; Setting the amplitude axis resolution of the phase resolution partial discharge map to be 256, and constructing a two-dimensional counting matrix, wherein the phase axis resolution is 256; Mapping the amplitude of each discharge pulse event into an amplitude index, and mapping the power frequency phase of each discharge pulse event into a phase index; for each discharge pulse event, accumulating and counting the corresponding element positions in the two-dimensional counting matrix according to the amplitude index and the phase index, and forming a phase resolution partial discharge map by taking the numerical value of each element in the two-dimensional counting matrix as the image pixel intensity.
3. The partial discharge pattern recognition method based on the phase-independent characteristics of the uhf signal according to claim 1, wherein the preprocessing phase-resolved partial discharge pattern comprises: performing self-adaptive removal processing of the bottom noise threshold based on the two-dimensional counting matrix to obtain a two-dimensional counting matrix after the self-adaptive removal processing of the bottom noise threshold; Performing phase resolution remapping on the two-dimensional counting matrix subjected to the self-adaptive removal processing of the bottom noise threshold to form a two-dimensional counting matrix subjected to phase remapping; Pulse number normalization processing, amplitude normalization processing and energy normalization processing are carried out on the two-dimensional counting matrix after phase remapping, and a two-dimensional counting matrix after normalization processing is obtained; and resampling the amplitude axis of the normalized two-dimensional counting matrix according to the uniform resolution to obtain a preprocessing phase resolution partial discharge map with uniform size.
4. The partial discharge pattern recognition method based on the phase independent characteristics of the uhf signal according to claim 1, wherein the splicing according to a preset sequence forms a phase independent composite characteristic vector, and the method comprises the following steps: Extracting an amplitude-phase statistic feature vector based on the preprocessing phase resolution partial discharge map; extracting pulse number-phase statistical feature vectors based on the preprocessing phase resolution partial discharge map; extracting pulse number-amplitude statistical feature vectors based on the preprocessing phase resolution partial discharge map; extracting an energy-phase statistical feature vector based on the preprocessing phase resolution partial discharge map; Extracting a frequency domain related feature vector based on the amplitude-phase statistical feature vector; and connecting the amplitude-phase statistical feature vector, the pulse number-amplitude statistical feature vector and the energy-phase statistical feature vector with the frequency domain related feature vector in a head-tail mode according to a preset sequence to obtain the phase irrelevant composite feature vector.
5. The partial discharge pattern recognition method based on the phase independent feature of the uhf signal according to claim 1, wherein the performing the adaptive feature dimension reduction process based on the class-separable contribution degree on the phase independent composite feature vector includes: constructing a training data set based on the phase-independent composite feature vector; splitting each phase irrelevant composite feature vector in the training data set one by one according to feature components, and respectively collecting values of all training samples in the feature dimensions of each category and each feature dimension to form a value set of each category in each feature dimension; for each feature dimension, calculating the intra-class variance of all the classes on the feature dimension, and accumulating the intra-class variances of all the classes on the feature dimension to obtain the total intra-class dispersion of the feature dimension; for each feature dimension, calculating a global mean value and calculating the inter-class dispersion of the ith feature dimension; Calculating a discrimination contribution factor for each feature dimension; Sorting the discrimination contribution factors of all the feature dimensions from large to small, sequentially accumulating the discrimination contribution factors of the first feature dimensions after sorting, and calculating the accumulated contribution rate, wherein the number of the selected feature dimensions is the feature dimension which is finally reserved when the accumulated contribution rate reaches a preset threshold value; and sequentially extracting feature components of the corresponding sequencing rank in the phase-independent composite feature vector according to the finally reserved feature dimension to form a low-redundancy phase-independent distinguishing feature vector.
6. The partial discharge pattern recognition method based on the phase independent characteristics of the uhf signal according to claim 1, wherein the discriminating energy potential function model comprises: based on the low redundancy phase irrelevant discrimination feature vector, constructing a training sample required by discriminating an energy potential function model; calculating a sample confidence weight for each training sample in each category; According to the sample confidence weight, carrying out weighted summation on low redundancy phase irrelevant discrimination feature vectors of all training samples in the same category to obtain a discrimination prototype center of a corresponding category partial discharge mode; Based on the sample confidence weight and the discrimination prototype center, calculating a discrimination diffusion matrix under each category, and carrying out structural contraction treatment on the discrimination diffusion matrix to obtain a discrimination diffusion matrix after structural contraction; Constructing a discrimination contribution weighting matrix according to the characteristic dimension discrimination contribution factor, carrying out weighted summation on the discrimination diffusion matrix after structured shrinkage, the discrimination contribution weighting matrix and the identity matrix, and then carrying out inversion to obtain a stabilized inverse diffusion matrix of the corresponding category partial discharge mode; based on the center of the discriminant prototype and the stabilized inverse diffusion matrix, establishing a discriminant energy potential function corresponding to each class; and forming a discrimination energy potential function set by the discrimination prototype center, the stabilized inverse diffusion matrix and the discrimination energy potential function of each class of partial discharge modes, and taking the discrimination energy potential function set as a discrimination energy potential function model.
7. The partial discharge pattern recognition method based on the phase independence characteristic of the uhf signal according to claim 1, wherein the determining the partial discharge pattern to which the partial discharge sample to be measured belongs according to the minimum discrimination energy criterion comprises: Obtaining a low redundancy phase irrelevant discrimination feature vector of a partial discharge sample to be detected; calculating a discrimination energy potential function value for each partial discharge mode category according to the discrimination energy potential function calculation form; and forming a discrimination energy potential function value set by the discrimination energy potential function values calculated under all the partial discharge mode types, determining the index of the partial basis partial discharge mode type to which the partial discharge sample to be detected belongs based on the minimum discrimination energy criterion, extracting the corresponding definition rule of the partial discharge mode type from the pre-established corresponding relation table of the partial discharge mode type and the physical defect type, and classifying the partial discharge sample to be detected according to the partial discharge generation mechanism, the discharge pulse statistical characteristic and the phase irrelevant characteristic distribution characteristic.
8. The partial discharge pattern recognition method based on the phase independent characteristics of the uhf signal according to claim 7, wherein the pattern classification comprises: The point discharge mode is that when the center of a discrimination prototype corresponding to the partial discharge mode category index to which the partial discharge sample to be tested belongs is shown as discharge pulse amplitude concentration, the amplitude-phase statistical characteristic is in unimodal distribution, the duty ratio of the pulse number-amplitude statistical characteristic in a high-amplitude interval exceeds a first threshold, and the duty ratio of a low-frequency component in the frequency domain related characteristic exceeds a second threshold; A floating electrode discharge mode, wherein when the center of a discrimination prototype corresponding to a partial discharge mode category index to which a partial discharge sample to be detected belongs is shown that a discharge pulse amplitude distribution range is larger than a preset range, pulse number-phase statistical characteristics are distributed in a dispersed mode in a plurality of phase intervals, the energy-phase statistical characteristics have asymmetry, and medium-low frequency components in frequency domain related characteristics coexist; An insulation defect discharge mode, wherein when a discrimination prototype center corresponding to a partial discharge mode category index to which a partial discharge sample to be detected belongs shows that discharge pulse energy accumulation exceeds a third threshold value, an energy-phase statistical characteristic shows an enhancement trend in a specific phase interval, pulse number-amplitude statistical characteristics form stable distribution in a medium amplitude interval, and frequency domain related characteristics have periodic characteristics; And the free particle discharge mode is that when the center of the discrimination prototype corresponding to the partial discharge mode category index to which the partial discharge sample to be detected belongs shows that the distribution of the discharge pulse amplitude and the pulse number is discrete, the pulse number-phase statistical characteristic shows random dispersion characteristic, the fluctuation of the energy-phase statistical characteristic is higher than preset fluctuation, and the high-frequency component duty ratio in the frequency domain related characteristic exceeds a fourth threshold value.

Description

Partial discharge mode identification method based on ultrahigh frequency signal phase irrelevant features Technical Field The invention relates to the technical field of partial discharge, in particular to a partial discharge mode identification method based on the phase independence characteristic of an ultrahigh frequency signal. Background Partial discharge is used as an important precursor of GIS internal insulation defect, early identification of the partial discharge has very important engineering significance for preventing electrical accidents and guaranteeing safe and stable operation of a power grid, and an ultra-high frequency (UHF) detection technology has become one of the main technical means of the current GIS partial discharge monitoring due to the advantages of strong anti-interference capability, high sensitivity and suitability for on-line monitoring. The conventional UHF discharge identification method generally relies on the distribution characteristics of 'discharge amplitude-power frequency phase' reflected by a PRPD (pulse-duration) map, and realizes discharge pattern identification by analyzing the statistical rules of discharge pulses in different phase intervals. However, in the actual engineering field, because the GIS equipment has complex operation environment, the power frequency synchronous signals in the acquisition link are easy to be interfered, the voltage transformer has inherent phase errors, and meanwhile, the clock synchronous precision between the detection devices is limited, so that the corresponding power frequency phase information in the PRPD map is always inevitably drifted or even lost, on the other hand, the field partial discharge is always characterized by 'sporadic, randomness, sparsity', and the like, the pulse quantity is limited in a short time, and stable and complete statistical map features are difficult to form, thereby seriously reducing the identification accuracy based on the phase related features, and obviously restricting the robustness and universality of the model. The prior art still has the following outstanding problems that the phase dependence on power frequency is strong, when the phase synchronization error reaches millisecond level, the characteristic distribution of the PRPD map is obviously distorted, the characteristic boundary between different types of partial discharge becomes fuzzy, misjudgment is easy to generate, the adaptability to the even-type discharge pulse is insufficient, the conventional PRPD depends on long-time statistics, when the discharge presents low repetition rate characteristics, the short-time PRPD is difficult to form a discharge mode with distinguished level, the interference sources are abundant under complex working conditions, various electromagnetic interferences such as communication equipment, power electronic devices, radio emission sources and the like are commonly present on the site of a transformer substation, and pulse signals generated in UHF frequency bands can overlap with real partial discharge signals on amplitude-phase distribution, so that false triggering and false recognition are caused. With the development of GIS equipment to higher voltage class and larger capacity, the internal structure of the GIS equipment is more complex, the electromagnetic radiation propagation path and the spectrum structure generated by partial discharge show stronger nonlinearity and diversity characteristics, and the traditional identification method relying on single-phase information is difficult to meet the actual requirement of high-reliability operation. Disclosure of Invention The invention aims to provide a partial discharge mode identification method based on the phase independence characteristic of an ultrahigh frequency signal, which can remarkably reduce the interference of noise characteristics and weak discrimination characteristics on a model and improve the stability and robustness of the discrimination model under the conditions of small samples and short-time monitoring. According to the embodiment of the invention, the partial discharge mode identification method based on the phase independence characteristic of the ultrahigh frequency signal comprises the following steps: Collecting an ultrahigh frequency time domain pulse signal by using a UHF receiving device, mapping the ultrahigh frequency time domain pulse signal into amplitude-power frequency phase two-dimensional coordinate data, and generating a phase resolution partial discharge map; Preprocessing the phase resolution partial discharge spectrum, and resampling according to the uniform resolution to obtain a preprocessed phase resolution partial discharge spectrum with uniform size; Extracting an amplitude-phase statistical feature vector, a pulse number-amplitude statistical feature vector, an energy-phase statistical feature vector and a frequency domain related feature vector based on the preprocessing phase resolution parti