Fuzzy Self-Organizing Maps for detection of Partial Discharge signals

Partial discharge (PD) detection has been used in assessment of condition reliability of electrical insulation in high voltage equipment such as power station. Unfortunately, PD signals took during condition monitoring are often corrupted with excessive interference. The challenge to effectively and accurately determine and extract the pure PD signal from the large amount of noise still remains. The focus of this paper is to explore artificial intelligence as a new denoising method for pure PD signal detection, especially for extracting low amplitude PD signals that are initially grouped with the noise signals. A fuzzy self-organizing maps (FSOM) method has been developed. It combines the concepts of Kohonen self-organizing maps (SOM) with fuzzy sets theory. A fuzzy classifier based on the FSOM is built to eliminate noise and extract pure PD signals. Two sets of laboratory-simulated signal data, surface and cavity, were used for the method verification. It is shown that the developed fuzzy classifier is superior to conventional threshold-filtering method in extracting the PD signals in the lower amplitude range.