A New Method for Partical Discharge Pattern Recognition of Electrical Transformers

Ultrasonic techniques are wildly used in online partial discharge (PD) location and recognition for electrical transformers. This paper focuses on a new ultrasonic feature extraction method. The normalized discharge grey moment features are extracted from ultrasonic signals to perform PD recognition. These features are sent to an improved BP neural network as to perform pattern recognition. Two types of PD patterns, pin-plate discharge and sphere-plate discharge, are tested, the PD pattern recognition method are compared with traditional methods and the recognition rates show that central grey moment has satisfactory ability in characterizing PD types, and the improved back propagation (BP) neural network perfectly met the recognition demands. Central grey moment provides us a novel approach to study partial discharge.