A Method for the Approximate Location of High Impedance Faults using Neural Networks

This paper describes an algorithm for ground to earth high-impedance fault identification and location in medium-voltage of unbalanced distribution systems. The daily variation of the load is considered, the fault resistance variation, inaccuracy in feeders data, and errors from phasor measurement units. The method is based on artificial neural networks associated with phasor measurement units to detecting the faulty area of the distribution system. Three types of neural network structures are proposed, (i) using phasor inputs, (ii) using non-phasor inputs, and (iii) using inputs to sequence components. To test the proposed method, thousands conditions of normal operation and fault operation were simulated to assemble the database. The method uses the current data obtained from meters that can be allocated at different points of the distribution system. The IEEE 123 test system is used to validate the proposed algorithm.