Classification of Partial Discharge Fault Sources on SF₆ Insulated Switchgear Based on Twelve By-Product Gases Random Forest Pattern Recognition

Abstract

Sulphur hexafluoride (SF6) gas insulated switchgear (GIS) is widely used in electrical power supply system and therefore needs regular preventive maintenance. Prediction and diagnosis analysis of faults in GIS using SF6 gas byproducts was introduced previously by using 4 to 8 types of by product gases. As latest development on gas analyser, more by-product gases can be detected and used for condition monitoring of the GIS. The type, number, concentration and chemical stability of by-product gases of SF6 GIS are found to be closely correlated to the type of defect. However, the number of byproduct gases used increases, the pattern for faults classification become more complex. Thus, further analysis on increasing number of by product gases using intelligent techniques such as pattern recognition is required. In this paper, 12 significant by-products captured due to various sources of partial discharge fault in GIS were used. Random Forest (RF) was selected in this work as a multi-class classification technique. The analyses using RF pattern recognition with eight algorithms based on the presence and concentration of the gas by-products were carried out. The RF algorithm successfully recognises a given defect with an accuracy of 87.5% for all defects fault classification. The performance of the RF algorithm is 1.5 times better than the decision table algorithm which is the next best algorithm. This research illustrates the feasibility and applicability of an effective GIS diagnostic using gas by-products analyses, and in particular, using the RF pattern recognition.

Subdomain Adaptation Capsule Network for Partial Discharge Diagnosis in Gas-Insulated Switchgear

“…In the process of forward propagation, cij is obtained using Equation ( 7) and vj is received according to Equations ( 5) and (6). cij is updated and modified utilizing the iteration of bij, and bij is from the change in vj.…”

“…Feature extraction uses signal processing technology, such as wavelet packet decomposition [2] and the short-time Fourier transform [3], to denoise and extract representative features. PD type classification utilizes different classification methods such as support vector machines [4] and K-nearest neighbor [5] and random forest [6] approaches. However, although manual feature extraction in ML methods seriously relies on expert experience, the performance of the classifier is greatly affected by the feature and generalization ability of the ML model; thus, there are great discrepancies among different classifiers under different states.…”

SF6Decomposed Component Analysis for Partial Discharge Diagnosis in GIS: A Review

“…Meanwhile, SF6 decomposition under three categories of PD sources was studied [86]. In the research, the random forest algorithm was employed.…”

Pattern Recognition of Partial Discharge Faults in Switchgear Using a Back Propagation Neural Network Optimized by an Improved Mantis Search Algorithm

“…The PDPRs are typically based on the signal characteristics generated in PD events, which can be classified into two major categories: chemical signals and physical signals. Chemical signals are mainly various gas derivatives produced in PD [1][2][3][4]. Muhamad et al, employed the random forest (RF) algorithm to conduct the pattern recognition of a gas insulated switchgear (GIS) using 12 by-product gases, and the performance of RF confirmed the feasibility of eight algorithms [3].…”