Waveform Analysis-Based Intelligent Fault Classifier with Novel Fault-Zone Segregations for Real-Time Application in Large Power Networks

Abstract

The article proposes a power system fault-recognition technique founded on waveform analysis to identify and categorize short circuit faults in large power systems. The main criticism against waveform-based approaches is their huge computational burden and relatively slow processing, making them unsuitable for on-line applications. This article efficiently handles the computational load by applying a novel zone partitioning concept, where a large power system is partitioned, forming smaller hypothetical zones with fewer number of buses and lines. Crude determination of fault zone is first performed by a neural network (NN)-based zone classifier. A succeeding module of classifier, then determines the exact fault nature and its precise location within a hypothetical zone. Transient recorders installed at optimal monitoring buses, in each partitioned zone, collect the preliminary wave-data, which is then passed-on to an extended Kalman filter (EKF)-based feature extraction module, to filter-out relevant fault features. The EKF-NN combination is found to be highly efficient and precise in determining fault nature and location when tested on WECC-200-bus practical system. Hypothetical-zone partitioning has been established as a highly efficient technique, reducing the computation burden and CPU usage considerably, and rendering the scheme suitable for on-line implementation. The method also improved the fault-recognition accuracy phenomenally.