The tim ely detection and isolation of High Impedance Faults(HIFs) are im portan t to ensure the safety of the public and the property around power distribution lines. The existing protection measures are highly inadequate for this task. Although a num ber of attem pts to develop HIF detectors were m ade, all of them were unable to provide a complete solution to this problem. The inability of conventional pattern recognition approaches to classify noisy spatio-tem poral patterns in environments such as, inexact detection knowledge and m ulti-param eter relationships, is considered to be the prim ary reason for the status quo. The objective of this research work is to explore the use of Artificial Neural Network(ANN) techniques in developing a HIF detection system. As the first step towards this ultim ate goal, the capabilities of three ANN architectures to detect these faults were investigated for this research work. The network architecture is critical to the success of any classifier developed using ANN approaches. This is so, because, the architecture dictates the network topology, the training algorithm to be used to train networks, and th e training procedure. Since HIF detection is a spatio-tem poral p attern recognition problem, the capabilities of three spatio-tem poral ANN approaches, buffered Multi-Layer Perceptron(M LP), Time-Delay Neural Network(TDNN), and Simple Recurrent Network(SRN), were investigated for this research. This dissertation describes the work we have done to explore the use of the three ANN architectures in detecting HIFs. Two of those architectures, TDNN and SRN, were applied outside of their prim ary operational domains and paradigms. The details of the new m odular feature extraction process th at was developed to facilitate HIF detection, the issues involved in training networks from the three ANN architectures, the output performances of the developed networks are described in this dissertation. Additionally, the network analysis work th at was done to determ ine the characteristics of the detection procedures used by the networks are also described.
