A novel fault location method for distribution systems using phase-angle jumps based on neural networks

This work presents a fault location method in distribution systems based on neural networks using a phase-angle jump as the model's single input. The IEEE 34 nodes system was used. Different fault scenarios have been considered to train ANN models including various incipient angles, fault types, fault resistance values, and various fault distances that typically affect a fault location algorithm's accuracy. Different load conditions were not considered in this particular study. Nine different models were trained specifically with particular fault types and fault resistance values and one model was trained with all fault scenarios regardless of the latter obtaining the best performance with three different models trained specifically for locating each fault type considered.