CN-121980226-A - Cable partial discharge defect identification method and system based on multichannel neural network

CN121980226ACN 121980226 ACN121980226 ACN 121980226ACN-121980226-A

Abstract

The invention relates to the technical field of power data processing, in particular to a cable partial discharge defect identification method and system based on a multichannel neural network. The cable multichannel mixed signals are collected through multiple sensors, a separation signal matrix and a mixing matrix are obtained through blind source separation, a scaling coefficient matrix is calculated through a self-adaptive complex calibration decision and a reference channel calibration method, amplitude correction is carried out, polarity and sequencing correction is completed subsequently, a physical quantity recovery signal is obtained, the physical quantity recovery signal is input into a pretrained multichannel neural network, and then the defect type can be identified, the discharge severity can be estimated, and the accuracy, reliability and quantitative analysis capability of a cable partial discharge defect identification technology are greatly improved.

Inventors

LI ZIKANG
WANG ZIYI
LIU HAORAN
ZHOU HUAYANRAN
LU XIN
ZHANG JIFENG

Assignees

国网上海市电力公司

Dates

Publication Date: 20260505
Application Date: 20260407

Claims (10)

1. The cable partial discharge defect identification method based on the multichannel neural network is characterized by comprising the following steps of: The method comprises the steps of acquiring multichannel mixed signals of a target cable simultaneously based on a plurality of sensors deployed at preset key positions of the cable, performing blind source separation on the multichannel mixed signals to obtain a separation signal matrix and a mixing matrix, performing self-adaptive complex calibration decision based on the separation signal matrix and the mixing matrix to obtain a complex calibration execution instruction, and calculating a scaling coefficient matrix based on a reference channel calibration method when the complex calibration execution instruction indicates to execute amplitude complex calibration; Performing amplitude correction on the separation signal matrix based on the scaling coefficient matrix to obtain an amplitude corrected signal, performing polarity correction and sequencing correction on the amplitude corrected signal to obtain a physical quantity recovery signal, inputting the physical quantity recovery signal into a pre-trained multichannel neural network, and performing defect recognition to obtain predicted fault data.
2. The method for identifying partial discharge defects of a cable based on a multichannel neural network according to claim 1, wherein simultaneously acquiring multichannel mixed signals of a target cable based on a plurality of sensors disposed at preset key parts of the cable comprises: Arranging a plurality of sensors at preset key positions of a preset cable, wherein the plurality of sensors comprise at least two of a high-frequency current transformer, an ultrahigh-frequency sensor and an ultrasonic sensor; synchronously collecting output signals of the plurality of sensors to obtain an original multi-channel signal; And performing filtering denoising processing on the original multichannel signal to obtain the multichannel mixed signal.
3. The method for identifying cable partial discharge defects based on a multichannel neural network according to claim 1, wherein performing an adaptive complex calibration decision based on the separation signal matrix and the mixing matrix to obtain a complex calibration execution instruction comprises: calculating an inter-channel correlation coefficient between any two separated signals in the separated signal matrix; Calculating the signal-to-noise ratio of each separated signal in the separated signal matrix; calculating the pulse amplitude variation coefficient of each separated signal in the separated signal matrix in a plurality of continuous power frequency periods; Calculating a reconstruction error based on the mixing matrix, wherein the reconstruction error is a root mean square error between an original multichannel mixed signal and a signal reconstructed by using the mixing matrix and the separation signal matrix; When the inter-channel correlation coefficient is smaller than a first preset threshold, the signal-to-noise ratio is larger than a second preset threshold, the pulse amplitude variation coefficient is smaller than a third preset threshold and the reconstruction error is smaller than a fourth preset threshold, generating a complex calibration execution instruction for executing amplitude complex calibration; otherwise, generating a prompt instruction for re-acquiring the signal.
4. The method for identifying partial discharge defects of a cable based on a multi-channel neural network according to claim 1, wherein when the complex calibration execution instruction instructs execution of amplitude complex calibration, calculating a scaling coefficient matrix based on a reference channel calibration method comprises: Selecting at least one calibrated sensor from the plurality of sensors, and setting the sensor as a reference sensor, wherein the original signal amplitude of the reference sensor has a known physical quantity corresponding relation; Based on the mixing matrix, analyzing weight distribution of each discharge source on each sensor channel, extracting a discharge source which corresponds to the channel where the reference sensor is located and has the largest weight, and setting the discharge source as a target separation signal; Calculating the correlation coefficient of the original signal of the reference sensor and the target separation signal, and verifying the matching correctness; When the correlation coefficient is larger than or equal to a preset matching threshold value, judging that the matching is correct, calculating a scaling coefficient based on the ratio of the original signal amplitude of the reference sensor to the target separation signal amplitude, and constructing the scaling coefficient matrix based on the scaling coefficient; When the correlation coefficient is smaller than the preset matching threshold, judging that matching fails, re-extracting a discharge source which corresponds to a channel where the reference sensor is located and has a large weight from the rest discharge sources, setting the discharge source as a new target separation signal, and repeatedly executing a correlation coefficient verification step until a discharge source which is matched correctly is found or all discharge sources are traversed; And when the correlation coefficient which is larger than or equal to the preset matching threshold value still cannot be obtained after all the discharge sources are traversed, generating a reference channel calibration failure prompt, and switching to a standard pulse injection method to calculate a scaling coefficient matrix.
5. The method for identifying partial discharge defects of a cable based on a multi-channel neural network according to claim 1, wherein performing polarity correction based on the amplitude corrected signal to obtain a polarity corrected signal comprises: Loading a power frequency phase reference signal; calculating a cross-correlation function of the amplitude corrected signal and the power frequency phase reference signal; When the correlation coefficient corresponding to the maximum value of the cross correlation function is a negative value, judging that the amplitude corrected signal has polarity inversion, and multiplying the amplitude corrected signal by-1 to obtain the polarity corrected signal; And when the correlation coefficient is a positive value, directly taking the amplitude corrected signal as the polarity corrected signal.
6. The method for identifying a partial discharge defect of a cable based on a multi-channel neural network according to claim 1, wherein performing polarity correction and sequencing correction based on the amplitude corrected signal, obtaining a physical quantity recovery signal, comprises: Extracting stable characteristics of each separated signal in the polarity corrected signal, wherein the stable characteristics comprise at least one of discharge phase distribution characteristics, pulse rise time and energy concentration; Matching the stable characteristics with a pre-stored defect characteristic fingerprint library, and distributing a fixed channel identifier for each separation signal; And sequencing the signals after the polarity correction based on the channel identification to obtain the physical quantity recovery signal.
7. The method for identifying partial discharge defects of a cable based on a multichannel neural network according to claim 1, further comprising: When the self-adaptive complex calibration decision obtains a complex calibration execution instruction which does not execute amplitude complex calibration, the polarity correction and the sequencing correction are directly executed on the separation signal matrix, and the physical quantity recovery signal is obtained.
8. The method for identifying cable partial discharge defects based on a multi-channel neural network according to claim 1, wherein inputting the physical quantity recovery signal into a pre-trained multi-channel neural network, performing defect identification, and obtaining predicted fault data, comprises: loading a pre-trained defect identification model, wherein the defect identification model is a mixed network of a convolutional neural network and a long-term and short-term memory network; Inputting the physical quantity recovery signal into the defect identification model, and outputting defect type and confidence; and calculating the discharge severity based on the physical quantity recovery signal.
9. The method for identifying a partial discharge defect of a cable based on a multi-channel neural network according to claim 8, wherein calculating the discharge severity based on the physical quantity recovery signal comprises: extracting pulse amplitude sequences of all discharge sources in the physical quantity recovery signal, and executing peak detection on the pulse amplitude sequences to obtain the maximum amplitude of discharge pulses; calculating apparent discharge amount based on the maximum amplitude and a preset sensor calibration coefficient; Extracting a power frequency voltage phase signal synchronously acquired with the physical quantity recovery signal, and calculating discharge energy based on the apparent discharge quantity and an instantaneous voltage value of the power frequency voltage phase signal; the discharge severity is evaluated based on the apparent discharge amount and the discharge energy.
10. A system for identifying cable partial discharge defects based on a multichannel neural network, characterized in that the system is used for realizing the cable partial discharge defect identification method based on the multichannel neural network as claimed in any one of claims 1-9, and the system comprises: the multi-channel mixed signal acquisition module is used for simultaneously acquiring multi-channel mixed signals of the target cable based on a plurality of sensors deployed at preset key positions of the cable; the blind source separation execution module is used for executing blind source separation on the multichannel mixed signals to obtain a separation signal matrix and a mixed matrix, wherein the separation signal matrix comprises a plurality of independent discharge source signals, and the mixed matrix represents the contribution weight of each discharge source to each sensor channel; The complex calibration executing instruction obtaining module is used for executing self-adaptive complex calibration decision based on the separation signal matrix and the mixing matrix to obtain a complex calibration executing instruction; The scaling factor matrix acquisition module is used for calculating a scaling factor matrix based on a reference channel calibration method when the complex calibration execution instruction indicates to execute amplitude complex calibration, wherein when the scaling factor matrix is calculated, the calculation priority and the matching relation of the scaling factors of each channel are determined based on the weight distribution of each discharge source on the sensor channel reflected by the mixing matrix; the amplitude corrected signal acquisition module is used for executing amplitude correction on the separation signal matrix based on the scaling coefficient matrix to obtain an amplitude corrected signal; the physical quantity recovery signal acquisition module is used for executing polarity correction and sequencing correction based on the amplitude corrected signal to obtain a physical quantity recovery signal; The predicted fault data acquisition module is used for inputting the physical quantity recovery signal into the pre-trained multichannel neural network, performing defect identification, and obtaining predicted fault data, wherein the predicted fault data at least comprises defect types and discharge severity.

Description

Cable partial discharge defect identification method and system based on multichannel neural network Technical Field The invention relates to the technical field of power data processing, in particular to a cable partial discharge defect identification method and system based on a multichannel neural network. Background In cable partial discharge detection, signals synchronously collected by multiple sensors are often a mixture of multiple discharge sources and noise. Blind source separation techniques are widely used for signal preprocessing because of the ability to separate independent discharge sources from the mixed signal. However, the blind source separation algorithm has inherent problems that the amplitude scaling uncertainty causes the separation signal to lose the original physical magnitude, the polarity inversion uncertainty distorts the discharge phase characteristics, and the ordering uncertainty causes the separation result to not correspond to the discharge source stability. In order to solve the above problems, the prior art, for example, patent application number CN201811211078.2, proposes a method for separating blind sources of convolution in noisy domain, which estimates noise variance by using a thin plate spline model and performs depolarization treatment in the frequency domain, so as to improve the separation precision under low signal-to-noise ratio, but only serve to correctly restore the waveform in amplitude correction, and fail to restore the physical magnitude of the signal for subsequent quantitative analysis. The patent with the application number of CN202511274080.4 further introduces an amplitude complex calibration concept, and calculates a scaling factor by using standard pulse injection, but the complex calibration decision of the method depends on whether deep learning denoising is introduced or not, and the problems of polarity inversion and sequencing uncertainty are not solved by a system, so that the signal after complex calibration is still difficult to directly use for quantitative defect diagnosis based on the amplitude. Therefore, the inherent amplitude scaling, polarity inversion and sequencing uncertainty of blind source separation are difficult to completely eliminate in the prior art, so that the physical magnitude of the separated signals is lost, and effective adaptation cannot be realized with a neural network, thereby restricting the accuracy and quantitative analysis capability of the cable partial discharge defect identification technology. Disclosure of Invention Aiming at the problems that the inherent amplitude scaling, polarity inversion and sequencing uncertainty of blind source separation are difficult to completely eliminate in the prior art, the physical magnitude of separation signals is lost, and effective adaptation cannot be realized with a neural network, so that the accuracy and quantitative analysis capability of a cable partial discharge defect identification technology are restricted, the invention provides a cable partial discharge defect identification method and system based on a multichannel neural network to solve the problems. The technical scheme for solving the technical problems is as follows: in a first aspect, the present invention provides a method for identifying partial discharge defects of a cable based on a multichannel neural network, including: simultaneously acquiring multichannel mixed signals of a target cable based on a plurality of sensors deployed at preset key positions of the cable; Performing blind source separation on the multichannel mixed signals to obtain a separation signal matrix and a mixing matrix, wherein the separation signal matrix comprises a plurality of independent discharge source signals, and the mixing matrix represents the contribution weight of each discharge source to each sensor channel; based on the separation signal matrix and the mixing matrix, executing self-adaptive complex calibration decision to obtain a complex calibration executing instruction; When the complex calibration executing instruction indicates to execute amplitude complex calibration, calculating a scaling coefficient matrix based on a reference channel calibration method, wherein when calculating the scaling coefficient matrix, the calculating priority and the matching relation of the scaling coefficients of each channel are determined based on the weight distribution of each discharge source on the sensor channel reflected by the mixed matrix; performing amplitude correction on the separation signal matrix based on the scaling coefficient matrix to obtain an amplitude corrected signal; Performing polarity correction and sequencing correction based on the amplitude corrected signal to obtain a physical quantity recovery signal; Inputting the physical quantity recovery signal into a pre-trained multichannel neural network, and executing defect recognition to obtain predicted fault data, wherein the predicted fault da