Multiple Linear Regression Models for Partial Discharge Location Along Transformer Windings

Abstract

More recently, it has been noted an increase in the incidence of failures in power transformers, mainly due to degradation of the solid insulation system. The identification of current pulses related to Partial Discharges (P Ds) has enabled the early detection and location of the solid insulation deteriorated regions, which is very useful to optimize the planning of the maintenance interventions, preventing unexpected power interruptions and reducing downtime. Even though several techniques for P Ds location have been proposed in recent years, limitations can still be noted, such as the inability to specify the discharge location in narrower regions and unproved applicability in windings of different configurations. In view of these, in this work, a P D location method is analysed, based on multiple linear regression models developed in conjunction with stepwise variable selection and analysis of variance methods. The technique is applied from P Ds simulation data, obtained through accurate transformer high frequency modelling, using statistical and principal components features of the terminal windings current signals. The results show that the P Ds can be effectively located, with high accuracy and success rate, for both layer-type and disk-type windings of transformers.