CLASSIFICATION OF PARTIAL DISCHARGE SOURCES IN GAS-INSULATED SUBSTATIONS USING NOVEL PREPROCESSING STRATEGIES

Partial discharge activity in GIS can indicate impending breakdown. Th e activity is recorded as characteristic point on wave records which a re classified into defect types by expert engineers. Classification as sists with the assessment of the risk of catastrophic failure. An auto matic technique has been developed aimed at replicating the strategy u sed by an expert during manual classification. Preprocessing technique s have been selected which implicitly incorporate the processes used b y the expert rather than applying explicit rules. Information from the records pertinent to classification is extracted using mathematical m orphology techniques allowing substantial compression of the data. An artificial neural network trained with such data has been shown to pro duce results indicating that classification matched that of the expert .