Discharge monitoring system development and pattern recognition method study for power generators

This paper first introduces a computerized partial discharge monitoring system for generators and then presents the experimental and numerical study of discharge pattern recognition method The system has specially designed transducers, data acquisition unit and system software, can obtain statistical as well as individual information of discharges. In order to validate the performance of the system, model discharge experiments were done. Feature extraction of the gathered data and artificial neural network (ANN) classification of the model discharge patterns were studied. ANN training, testing and discharge pulse response simulation demonstrated that surface fitting method is suitable to extract features from statistical data of discharges, and ANN is a potential classifier that can be used in practice.