Evaluation of the impact of different neural network structure and data input on fault detection

This paper presents a methodology for detecting single-phase fault in medium voltage distribution systems using neural network. The impact of diverse types of neural network topologies and different inputs in the performance of the fault detection is assessed. The entries of the neural network are the electrical parameters of voltage and current, which could be phasor and / or non-phasor measurements. The proposed methodology aims to identify single-phase faults with an arc resistance ranging from 0 to 20 ohms, considering the variation of the system load.