Partial Discharge Fault Identification Algorithm Research Based On Ultrasonic Signals

The switchgear is one of the main equipment in the power system, partial discharge isone important reason to cause the switching cabinet insulation caused by the aging poweraccident. The online testing of insulated switchgear status and development trend, will beable to determine the insulation internal failure of the existence of local defects. The partialdischarge phenomenon will be accompanied by the physical process of the sound change,since in recent years, ultrasound based switchgear partial discharge on line detection has beenapplied as the power sector an important means to detect and diagnose insulation condition ofthe switchgear.We can make use of the audio signal processing and pattern recognition classificationalgorithm to identify the classification of the collected partial discharge ultrasonic signal, inorder to determine which devices exist insulation failure Drop frequency of the ultrasonicsignal processing to obtain a signal in the voice frequency range, extract the characteristicparameter of partial discharge signals, using voice recognition technology also can achievethe purpose of identification, and ultimately achieve the online non invasive detection of theinsulated state of the high voltage equipment.The main research contents can be concluded as follows:(1)According to the characteristics of audio,39audio features are extracted in timedomain and frequency domain.(2)Aiming at the problem of high dimensions and strong correlation in feature set, afeature optimization method based on Partial Least Squares is discussed. The optimizedfeature set includes fewer features and the new features are independent. The performances offeature optimizations based on Partial Least Squares (PLS) and Principal Component Analysis(PCA) are compared.(3)The hybrid approach based on PLS and Support Vector Machine (SVM) is proposedfor audio classification. This approach uses PLS to optimize feature set and then theoptimized feature set serves as the input of SVM. The experiment shows the classificationaccuracy is good.(4)The audio classification approach based on PLS SVM is applied to ultrasonic partial discharge detection system.