Real time high-impedance fault detection in power distribution system based on artificial neural network

High-impedance faults (HIFs) often occur when energized conductors contact with the objects of high resistance. HIFs in the initial stage have a low current magnitude, which is difficult to detect by conventional protection systems. Therefore, effective real-time fault detection technologies are required to diagnose HIFs and prevent damages in power distribution system in advance. This study proposes a methodology for HIFs detection in a real-time manner using artificial neural network (ANN). First, root mean square (RMS) is utilized to preprocess the high-dimensional current data and calculate RMS currents. Then, 67 time-domain features are obtained from RMS currents. The features are then trained by a multilayer perceptron network to build the ANN prediction model, and the model was optimized by the back-propagation algorithm. Grid search and cross-validation are utilized to search for the optimal values of parameters and obtain an objective result. The proposed methodology is applied on current data of 210 HIFs tests by Victorian Government. The results show that the proposed method can detect the HIFs with an accuracy of 95% in power distribution systems in a real-time manner. Compared with other machine learning methods, such as support vector machine and decision tree, the ANN based methodology have better performance on classification.