Method for Locating Single-Phase Disconnection Faults in a Smart Distribution System Based on Refined Composite Multiscale Permutation q-Complexity Entropy and Random Forest

In this article, a novel hybrid method of combining refined composite multiscale permutation${q}$-complexity entropy (RCMPQE) and random forest (RF) is developed to locate disconnection faults in a power distribution system. As a feature extraction method, RCMPQE is used to analyze the negative-sequence current to obtain fault features, and then they are input into RF to build a fault location model. The effectiveness of the proposed method is corroborated by referencing a three-feeder distribution network. The results demonstrate that compared with multiscale permutation entropy (MPE), refined composite MPE (RCMPE), and generalized MPE (GMPE), the RCMPQE has stronger feature extraction ability and can abstract more prominent fault information from negative-sequence current signals. The proposed method for locating faults significantly outperforms other commonly employed machine learning models, and it is immune to various levels of noise. Furthermore, the proposed method can perform well with a high fault location accuracy in a new system.