Identification of partial discharges sources using a combination of linear prediction and neural networks

Abstract

The ability to recognize forms by means of neural networks combined with a linear prediction analyze has been developed in this paper for partial discharges (PD) sources recognition. The PD sources are artificially created on a Teflon cell having two armatures isolated by polypropylene films. This experiments have been made at LEMD/CNRS Grenoble (France). This first work has been carried out by the use of a sample of seven specimens representing seven cases of presence of cavities on the cell insulation of study. The seven defects represented by respective signals of apparent charge were processed modeled by linear prediction and then passed to a back propagation neural networks.