Pattern Classification based Intelligent Numerical Protection of Salient - pole Synchronous Generator using Neural Networks

This paper presents the application of neural networks for the non-differential protection of salient-pole synchronous generator against internal faults in any winding of the stator. The direct phase quantities and modified winding function approach has been used to simulate different types of internal and external faults using electrical parameters of generators installed by utilities. The cases of all the possible types of internal faults in the stator winding have been taken into consideration in designing the protection scheme. Multi-layer Feed-forward Neural Network (MFNN) and Radial Basis Function Neural Network (RBFNN) have been trained and tested for detection, identification and classification of the internal faults based on pattern classification. The simulated fault currents in the phases as well as their parallel paths at the terminal end have been used for training and testing of both the proposed neural networks. Both the networks are able to identify the fault signal correctly but the MFNN is more reliable, more accurate and faster than RBFNN in detection and classification of the fault.