Detection and classification of line faults on power distribution systems using neural networks

This paper presents a new neural network approach based on a clustering algorithm to detect and classify line faults in a power distribution system. A robust features preprocessing procedure is discussed which extracts meaningful features from current wave forms to serve as a reduced set of inputs to the neural network. Lastly, results are given from studies that were conducted to determine the optimal order of presentation of the training feature patterns and the set of features that are necessary for the neural network to perform arcing identification.<
>