Identification of Partial Discharge Source in Power Apparatus in Practical Substation Utilizing Artificial Neural Network

摘要

In the previous report, two types of artificial neural network (ANN) were constructed based on waveform data and phase-resolved partial discharge (PRPD) patterns obtained from simulated defect samples in the laboratory. In this report, the two types of ANN were used to discriminate between PD and noise for signals acquired from a 72 kV tank type gas insulated switchgear (GIS) installed at a substation. The signals obtained from the GIS also include those caused by abnormalities due to floating electrodes that were confirmed by internal inspection after the measurement. It was found that the signals caused by the abnormality were mainly judged as noise and PD by ANN_WP and ANNJPR, respectively. The disagreement by ANN_WP is attributed to the waveform information change due to signal propagation and attenuation, while ANN_PR provides better judgement owing to keeping the phase information at which PD occur. As a result, ANN_PR is found to be more effective means to judge the PD source in the field.