A novel fault classification technique for double-circuit lines based on a combined unsupervised/supervised neural network

Thc work described in this papcr addresses the problems encountcrcd by conventional techniques in fault type classification in doublecircuit transmission lines; these arisc principally due to the mutual coupling between the two circuits under fault conditions, and this mutual coupling is highly variable in nature. It is shown that a neural network based on combined unsupervisdlsupcrvised training methodology provides the ability to accuratcly classify the fault type by identilying different patterns of thc associated voltagcs and currents. Tie technique Is compared with tlial based solely on a supervised training algorithm (ie back-propagation network classifier). It is then testcd under different fault types, location, resistance and inception anglc; different sourcc capacities and load angles are also considercd. All the teat mulls show that the proposed fault classifier is very well suited for classifying fault types in doublc-circuit lines.