Partial Discharge Analysis using PCA and SOM

Continuous on-line monitoring of partial discharge (PD) activities involves recording large amounts of data and a challenging task is to extract useful information from them. Thus, data mining plays a very important role. In this paper, data of PD in power cables are obtained from on-site monitoring and from laboratory test. The PD patterns in terms of the univariate phase-resolved distributions are analysed. An alternative method to that using statistical moments for characterizing the patterns is proposed. Principal component analysis (PCA) is used for feature extraction as well as dimensionality reduction in the proposed method. The results obtained are further processed using self-organizing mapping (SOM) to enable better visualisation of the trend of the PD activities.