A cascade perceptron and Kohonen network approach to fault location in rural distribution feeders

The location of faults in the distribution system is one of the most important problems for utilities in the sector now since they directly impact the quality of the energy supply, as well as economical. Thus, the faster the fault identification and the electrical energy prompt reestablishment, the better are the quality indicators of these companies. In this scenario, this work presents an alternative and efficient method that uses artificial neural networks to locate faults in distribution feeders. The developed methodology was applied using real data from a rural distribution system. Thus, through simulations in the ATPDraw software, it is possible to obtain current and voltage values of the studied feeder, these being the inputs of the developed neural networks. The neural networks were developed in the Matlab (R) software in two stages: first, a Perceptron Multiple Layers networks perform the classification of the types of faults. These, in sequence, are used to select the Kohonen Self-Organizing Maps employed in the second stage to the location of the feeder fuse switch that acted at the time of the fault. Simulation results from the real system are presented to validate the proposed methodology, making it possible to accurately estimate the equipment that operated when the interruption occurred, which provides greater speed in reestablishing the electricity supply in the affected region, increasing the quality indicators and reducing maintenance and regulatory costs of electrical utilities. (C) 2020 Elsevier B.V. All rights reserved.