Neural fault classifier for transmission line protection - a modular approach

Fault classification is an essential task for transmission line protection. Different forms of neural network are proposed for fault classification and all are of monolithic in structure. In this work modular concept is introduced to the neural network solution of the fault classification problem. The fault classification task is divided into number of subtasks where each subtask is accomplished by a neural network. For this each phase or ground is provided with a probabilistic neural network to indicate on its state of involvement with the fault. Such a network considers corresponding phase/ground voltage and current information as input and thereby the redundant inputs in conventional approaches are eliminated. The modular neural network approach is applied to two fault classification problems of transmission system and found to be accurate.