Multi-feature audio fingerprint fault diagnosis method and system for power equipment based on machine learning

Abstract

The invention provides a multi-feature audio fingerprint fault diagnosis method and system for power equipment based on machine learning. The multi-feature audio fingerprint fault diagnosis method forthe power equipment based on machine learning comprises the steps that training stage is carried out, specifically, training samples are preprocessed, then feature extraction is carried out, 12-dimensional MFCC parameters and 12-dimensional delta MFCC parameters sufficient to represent signal features are extracted as observation sequences; according to the characteristics of continuous mixed Gaussian HMM, the parameters of HMM models are initialized, and initial values of the parameters are stored; the initial values of the parameters are reestimated and trained according to the observationsequences; the parameters of the reestimated HMM model are saved, and the HMM models are established; and diagnosis stage is carried out, specifically, test samples are obtained, the features of the test samples are extracted, and the output probability of observed value sequences under each HMM model is calculated according to the established HMM models. According to the multi-feature audio fingerprint fault diagnosis method and system for the power equipment based on machine learning, whether the equipment is in an abnormal operation state or not can be judged through the changes of audio characteristics such as the timbre of sound, the size of volume and the level of frequency, and even the type and severity of the fault can be judged.

Description

technical field [0001] The present disclosure relates to the technical field of fault diagnosis, in particular to a machine learning-based multi-feature audio fingerprint fault diagnosis method and system for power equipment. Background technique [0002] The structure of power equipment is complex, and its operating conditions are relatively harsh, so the probability of failure in the system is relatively high. The investigation found that the failures caused by machinery accounted for about 70% of MF and MF failures. The so-called MF refers to the main failure, which causes one or more than one basic function of the power equipment to fail. MF refers to secondary failures, that is, other failures besides the main failure. In the survey of power equipment above 63kV, it was found that among all the main faults, the faults caused by mechanical operation accounted for 44%, and among all the secondary faults, the faults caused by mechanical operation accounted for 44%. 39.4...

AI Technical Summary

Problems solved by technology

[0005] 1. The indoor substation equipment fault judgment method based on audio feature extraction is to use the inspection robot to carry a two-channel audio acquisition module to collect the substation indoor equipment such as protection screens, monitoring screens, communication screens, battery cabinets, and background machines. The audio information of secondary equipment, the main signal is the discharge sound signal, not real-time online monitoring

[0006] 2. Portable high-voltage equipment fault location device and location method based on sound recognition. The main content of the patent is a portable audio collection device, which collects audio information and transmits it to the background, and judges according to the stored audio fault characteristics. There is no Clarify targeting method

Moreover, this electromagnetic circuit device is susceptible to electromagnetic interference in the operating environment of power equipment with a strong magnetic field. The background does not have self-learning and training capabilities, and the degree of intelligence is not high.

Images