Partial discharge discrimination

• Main Author: Hao, Liwei
• Published: University of Southampton 2008
• Subjects: 621.31
• Online Access: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.485541

This thesis details the application of machine based learning techniques to partial discharge (PD) discrimination. The learning machine, namely the support vector machine (SVM) has been assessed as a potential toot for PD source identification. High voltage cable sections and a power transformer bushing system have been used as experimental models in laboratory. Obtained results from different PD sources under controlled laboratory conditions were processed using proposed approaches such as phase resolved methods including two-dimensional ep-q ep-n . histograms and three-dimensional ep-q-n patterns, improved pulse sequence analysis (PSA), Fourier transform and wavelet analysis. Features were subsequently extracted from the pre-processed data and evaluated when used as the characteristic vectors for a clustering technique and SVM classification. A comprehensive automatic PD identification system has been developed and assessed. Some very encouraging results have been achieved by using SVM based identification. . • The use of optical transmission tedlniques on PD monitoring of power transformers has also been assessed. This thesis details the application of an electro-optic modulator (EOM) to generate transmission signals over polarization maintaining optjcal fibre from the measurement point to a remote control point. A data mining method to increase detection sensitivity for an optical phase modulator based PD monitoring system has been developed. By ap'plying a SVM to denoising, the operational reliability of an on-line condition monitoring system for high voltage transmission assets can be greatly improved.