Stationarity in Partial Discharge Time Series of Electrical Trees

Abstract

Electrical treeing is one of the main degradation mechanisms in high voltage polymeric insulation. These trees are hollow tubes that grow inside the insulation under the action of partial discharge (PD) activity; therefore, PD analysis is key to diagnosing polymeric insulation. Two traditional methods for PD analysis are Phase Resolved PD (PRPD) patterns and Pulse Sequence Analysis (PSA). More recently, the use of nonlinear time series analysis (NLTSA) for PD analysis was proposed. A key concept in NLTSA is stationarity; although its formal definition does not apply to experimental data, almost all methods of NLTSA require stationarity. This research uses the Cross Prediction Error (CPE) algorithm to analyze stationarity in PD time series from electrical trees. Their results are evaluated through PSA and PRPD patterns, and NLTSA. This paper shows that tools that consider the information of PD sequences, such as PSA and NLTSA, are more sensitive to stationarity. Therefore, these tools can better detect changes in PD dynamics, thus helping to improve polymeric insulation diagnosis. Furthermore, the results suggested that CPE algorithm could detect changes in PD dynamics since it allows a time-resolved evaluation of stationarity.