Real-Time Automatic Anomaly Detection Approach Designed for Electrified Railway Power System

An automatic and intelligent abnormal electrical process detection scheme is crucial for protecting the stability and power quality of an electrical power system and further, the operation of the future grid. This paper introduces the automatic monitoring system for electrified railway power system and designs a framework based on the convolution neural network for abnormal electrical process detection, integrating the data processing, feature extraction, and classification into one model. Then inception blocks are introduced as a kernel-wise approach to boost the performance. The data from the railway electrification system is applied to this scheme and receives a high performance of 97% abnormal electrical process recognition rate.