CN-117238315-B - Method and device for unsupervised detection of abnormal sound of switch on power distribution network column

CN117238315BCN 117238315 BCN117238315 BCN 117238315BCN-117238315-B

Abstract

The method and the device for unsupervised detection of abnormal sounds of the switch on the distribution network column are used for determining frequency domain features and time domain features corresponding to audio signals of the switch on the distribution network column, performing dimension reduction processing on the frequency domain features and the time domain features corresponding to the audio signals by adopting a robust principal component analysis method, determining target time domain features and target frequency domain features, inputting the target time domain features and the target frequency domain features into a multi-scale signal regulation self-encoder model and a NKNN-based proximity model to obtain a first detection result and a second detection result, inputting the target time domain features and the target frequency domain features into a multi-element Gaussian mixture model to obtain a third detection result when the first detection result and the second detection result are inconsistent, voting the first detection result, the second detection result and the third detection result based on a voting strategy, and determining the target detection result of the audio signals according to the voting result. Therefore, the detection precision of abnormal sounds of the switch on the distribution network column can be improved.

Inventors

LIN XIANG
SUN JIAN
LI KAIRAN
KE QINGPAI
Qiu Yangxin
FANG JIAN
DAI XIAOFENG
TONG RUI
TIAN YAN
ZHANG MIN
YANG FAN
LIU TONG
SHI XUNTAO

Assignees

广东电网有限责任公司广州供电局

Dates

Publication Date: 20260508
Application Date: 20231031

Claims (10)

1. An unsupervised detection method for abnormal sounds of switches on a power distribution network column, which is characterized by comprising the following steps: Determining frequency domain characteristics and time domain characteristics corresponding to audio signals of switches on a power distribution network column; performing feature dimension reduction processing on frequency domain features and time domain features corresponding to the audio signals by adopting a robust principal component analysis method, and selecting target time domain features and target frequency domain features from the time domain features and the frequency domain features subjected to the feature dimension reduction processing respectively; inputting the target time domain features and the target frequency domain features into a pre-established multi-scale signal conditioning self-encoder model and into a pre-established NKNN-based proximity model to obtain a first detection result of the multi-scale signal conditioning self-encoder model and a second detection result of the NKNN-based proximity model; when the first detection result and the second detection result are inconsistent, inputting the target time domain features and the target frequency domain features into a pre-established multi-element Gaussian mixture model to obtain a third detection result of the multi-element Gaussian mixture model; and voting the first detection result, the second detection result and the third detection result based on a preset voting strategy, and determining a target detection result of the audio signal according to the voting result.
2. The method for unsupervised detection of abnormal sounds of on-pole switch of power distribution network according to claim 1, wherein the step of determining the frequency domain characteristics and the time domain characteristics corresponding to the audio signal of the on-pole switch of the power distribution network comprises: acquiring an audio signal of a switch on the power distribution network column; denoising the audio signal, and performing audio framing processing on the denoised audio signal; Windowing the audio signal subjected to the audio framing treatment; And respectively extracting frequency domain features and time domain features of the audio signals subjected to windowing processing to obtain the frequency domain features and the time domain features corresponding to the switch audio signals on the power distribution network column.
3. The method for unsupervised detection of abnormal sounds of on-pole switches of a power distribution network according to claim 1, wherein the step of performing feature dimension reduction processing on frequency domain features and time domain features corresponding to the audio signal by using a robust principal component analysis method comprises the following steps: constructing a frequency domain original feature matrix of a frequency domain feature corresponding to the audio signal and a time domain original feature matrix of a time domain feature corresponding to the audio signal; Decomposing the frequency domain original feature matrix into a frequency domain low-rank component and a frequency domain sparse component, and decomposing the time domain original feature matrix into a time domain low-rank component and a time domain sparse component; And determining the frequency domain characteristics after the dimension reduction processing according to the frequency domain low-rank component and the frequency domain sparse component, and determining the time domain characteristics after the dimension reduction processing according to the time domain low-rank component and the time domain sparse component.
4. The method for unsupervised detection of abnormal sounds of on-pole switches of a power distribution network according to claim 1, wherein the step of selecting the target time domain feature and the target frequency domain feature from the time domain feature and the frequency domain feature after the feature dimension reduction process, respectively, comprises: Determining a plurality of time domain feature values corresponding to the time domain features after the dimension reduction processing, and determining a plurality of frequency domain feature values corresponding to the frequency domain features after the dimension reduction processing; According to the sequence from big to small of the values of the time domain feature values, the time domain feature after the dimension reduction processing corresponding to the first N time domain feature values is selected as the target time domain feature, and according to the sequence from big to small of the values of the frequency domain feature values, the frequency domain feature after the dimension reduction processing corresponding to the first M frequency domain feature values is selected as the target frequency domain feature, wherein N and M are positive integers larger than 1, and N and M are different.
5. The method for unsupervised detection of abnormal sounds of on-pole switch of power distribution network according to claim 1, wherein if the detection categories corresponding to the first detection result, the second detection result and the third detection result are audio normal or audio abnormal, the step of voting the first detection result, the second detection result and the third detection result based on a preset voting strategy and determining a target detection result of the audio signal according to the voting result comprises: Counting the number of votes corresponding to the detection category of the normal audio and the number of votes corresponding to the detection category of the abnormal audio according to the first detection result, the second detection result and the third detection result; and taking the detection category with the largest voting number as the target detection result.
6. The method for unsupervised detection of abnormal sounds of on-pole switch of power distribution network according to claim 1, wherein if the first detection result, the second detection result and the third detection result each include a probability corresponding to a detection category of normal audio and a probability corresponding to a detection category of abnormal audio, the step of voting the first detection result, the second detection result and the third detection result based on a preset voting strategy and determining a target detection result of the audio signal according to the voting result comprises: determining a first confidence level of the multi-scale signal conditioning self-encoder model, a second confidence level of the NKNN-based proximity model, and a third confidence level of the multi-element gaussian mixture model, respectively; For the detection category of the normal audio, determining the weighted probability corresponding to the detection category of the normal audio through weighted summation according to the first confidence coefficient, the second confidence coefficient and the third confidence coefficient, and the probability corresponding to the detection category in the first detection result, the probability corresponding to the detection category in the second detection result and the probability corresponding to the detection category in the third detection result; for the detection category of the audio abnormality, determining the weighted probability corresponding to the detection category of the audio abnormality through weighted summation according to the first confidence coefficient, the second confidence coefficient and the third confidence coefficient, and the probability corresponding to the detection category in the first detection result, the probability corresponding to the detection category in the second detection result and the probability corresponding to the detection category in the third detection result; And taking the detection category with larger weighted probability as the target detection result according to the weighted probability corresponding to the detection category with normal audio frequency and the weighted probability corresponding to the detection category with abnormal audio frequency.
7. The method for unsupervised detection of abnormal sounds of switches on a distribution network column according to any one of claims 1 to 6, further comprising: and when the first detection result and the second detection result are consistent, taking the first detection result or the second detection result as a target detection result of the audio signal.
8. An apparatus for the unsupervised detection of abnormal sounds of a switch on a distribution network column, said apparatus comprising: The characteristic determining module is used for determining frequency domain characteristics and time domain characteristics corresponding to the audio signals of the switches on the distribution network column; The target feature selection module is used for carrying out feature dimension reduction processing on the frequency domain features and the time domain features corresponding to the audio signals by adopting a robust principal component analysis method, and selecting target time domain features and target frequency domain features from the time domain features and the frequency domain features subjected to the feature dimension reduction processing respectively; The detection result acquisition module is used for inputting the target time domain characteristics and the target frequency domain characteristics into a pre-established multi-scale signal adjustment self-encoder model and inputting the target time domain characteristics and the target frequency domain characteristics into a pre-established NKNN-based proximity model to obtain a first detection result of the multi-scale signal adjustment self-encoder model and a second detection result of the NKNN-based proximity model; The target feature input module is used for inputting the target time domain features and the target frequency domain features into a pre-established multi-element Gaussian mixture model when the first detection result and the second detection result are inconsistent, so as to obtain a third detection result of the multi-element Gaussian mixture model; And the target detection result determining module is used for voting the first detection result, the second detection result and the third detection result based on a preset voting strategy, and determining the target detection result of the audio signal according to the voting result.
9. A storage medium having stored therein computer readable instructions which, when executed by one or more processors, cause the one or more processors to perform the steps of the method for unsupervised detection of abnormal sounds of switches on a distribution network pole as defined in any one of claims 1 to 7.
10. A computer device includes one or more processors and a memory; Stored in the memory are computer readable instructions which, when executed by the one or more processors, perform the steps of the method for unsupervised detection of switching anomalies on a distribution network post of any one of claims 1 to 7.

Description

Method and device for unsupervised detection of abnormal sound of switch on power distribution network column Technical Field The application relates to the technical field of power systems, in particular to an unsupervised detection method and device for abnormal sounds of switches on a power distribution network column. Background The rapid development of industry causes the electricity consumption of China to be continuously increased, and has higher requirements on the transmission of electric power. Regular operation maintenance and maintenance of a distribution line and equipment thereof are the basis for ensuring normal and safe operation of a distribution network, and distribution operators need to develop regular inspection to discover the condition of threatening normal and safe transportation of electric power, and the inspection content mainly comprises overhead lines, equipment and channels, cable lines and cable channels, distribution stations, switching stations and the like. The audio monitoring method has the advantages of high efficiency and low cost, and the fault detection early warning based on sound can give out a danger warning in advance, so that the machine is maintained in time. In the operation maintenance of distribution lines, the detection of the on-pole switching equipment comprises whether the switch emits abnormal sound or not, so that the monitoring of the abnormal sound of the on-pole switching equipment can be realized by means of computer methods such as an audio signal acquisition technology, pattern recognition, machine learning and the like, manpower and time can be saved, and the overhaul process is more efficient and convenient. The current abnormal sound detection steps include sound data collection, data preprocessing, feature extraction, supervised classification, or unsupervised classification. The extracted sound features have no unified standard, and the features of the time domain and the frequency domain reflect the characteristics of the audio from different angles, but too high feature dimension can cause problems of slow running speed, low accuracy and the like of the model, and it is necessary to comprehensively consider the characteristics of all the features and reduce the dimension of the features. Therefore, in the prior art, the collection difficulty of the column switch abnormal sound sample is high, the sample size is small, so that the applicability of a detection model based on supervised learning design is low, and finally, the judgment accuracy of whether the sound is abnormal is not high. Disclosure of Invention The present application aims to solve at least one of the above technical drawbacks, and in particular, to solve the technical drawback of the prior art that the accuracy of judging whether the sound is abnormal is not high. In a first aspect, the present application provides an unsupervised detection method for abnormal sounds of a switch on a distribution network column, where the method includes: Determining frequency domain characteristics and time domain characteristics corresponding to audio signals of switches on a power distribution network column; performing feature dimension reduction processing on frequency domain features and time domain features corresponding to the audio signals by adopting a robust principal component analysis method, and selecting target time domain features and target frequency domain features from the time domain features and the frequency domain features subjected to the feature dimension reduction processing respectively; inputting the target time domain features and the target frequency domain features into a pre-established multi-scale signal conditioning self-encoder model and into a pre-established NKNN-based proximity model to obtain a first detection result of the multi-scale signal conditioning self-encoder model and a second detection result of the NKNN-based proximity model; when the first detection result and the second detection result are inconsistent, inputting the target time domain features and the target frequency domain features into a pre-established multi-element Gaussian mixture model to obtain a third detection result of the multi-element Gaussian mixture model; and voting the first detection result, the second detection result and the third detection result based on a preset voting strategy, and determining a target detection result of the audio signal according to the voting result. In one embodiment, the step of determining the frequency domain characteristic and the time domain characteristic corresponding to the audio signal of the switch on the distribution network post includes: acquiring an audio signal of a switch on the power distribution network column; denoising the audio signal, and performing audio framing processing on the denoised audio signal; Windowing the audio signal subjected to the audio framing treatment; And respectively extracting frequency domain featu