Attention based Bi-LSTM for Power Line Partial Discharge Fault Detection

Abstract

Partial discharge is one of the faults in power distribution networks, especially in overhead lines with covered conductor of the medium voltage distribution networks, which may damage equipment and stop it’s functioning entirely. However, it is more difficult to detect partial discharge faults because of the different physical and chemical reactions and the same characterization caused by partial discharge faults, and the noise interference of the environment itself. By deploying distributed high-frequency sampling sensors to collecting three-phase voltage signal and using machine learning algorithms to detect the presence of partial discharge fault, can not only not interfere with the normal work of transmission lines to realize on-line detection of partial discharge fault, also can pass the upgrade in fault detection fault detection algorithm which can improve accuracy, therefore, it has become one of the main methods of partial discharge fault detection. In this paper, an attention mechanism based bidirectional long short term memory (Attention-LSTM) fault detection model is proposed, which can take full advantage of the bidirectional long neural network (Bi-LSTM) and attention mechanism, hence, no feature engineering expert is needed to solve the power line partial discharge fault detection issues. Experimental results show that the proposed model outperforms the existing methods in all selected performance metrics.