Method of Fault Self-Healing in Distribution Network and Deep Learning Under Cloud Edge Architecture

Abstract

At present, the distribution network fault self-healing method based on deep learning in smart grid work often has problems such as low accuracy and insufficient feature extraction ability. To overcome this, the authors propose a method of fault self-healing in a distribution network based on robot patrol and deep learning in a cloud edge architecture. Firstly, the data collected by the robot fault collection system is preprocessed by using one-hot coding and normalization methods to prevent data flooding. Secondly, they propose an improved bi-directional short-term memory (BiLSTM) fault location method which combines the advantages of both BiLSTM and attention mechanism, adjusts attention weight, filters, or weakens redundant information. Finally, the I-BiLSTM network and the U-BiLSTM network are trained, respectively, and the fault section can be accurately located based on the data of each node of the robot fault collection system topology. Experimental results show that this method has achieved accuracy scores of 0.928, 0.933, 0.948, and 0.942, respectively, in four fault types, namely single-phase grounding, two-phase grounding, phase-to-phase short circuit, and three-phase short circuit, which outperform those in previous literature. The proposed method is well suited for applications in smart grid work because of its desirable fault self-healing ability.