Recognition of fault transients using a probabilistic neural-network classifier

Summary form only given. This paper investigates the applicability of decision tree, hidden Markov model, and probabilistic neural-network(PNN) classification techniques to distinguish the transients originating from the faults from those originating from normal switching events. Current waveforms due to different types of events, such as faults, load switching, and capacitor bank switching were generated using a high-voltage transmission system simulated in PSCAD/EMTDC simulation software. Simulated transients were used to train and test the classifiers offline. The wavelet energies calculated using three-phase currents were used as input features for the classifiers. The results of the study showed the potential for developing a highly reliable transient classification system using the PNN technique. An online classification model for PNN was fully implemented in PSCAD/EMTDC. This model was extensively tested under different scenarios. The effects of the fault impedance, signal noise, current-transformer saturation, and arcing faults were investigated. Finally, the operation of the classifier was verified using actual recorded waveforms obtained from a high-voltage transmission system.