A Non-intrusive Anomaly Detection Method for Distribution Integration Terminal

Abstract

In recent years, with the accelerated development of power IoT construction, the edge intelligent terminal devices have been developed and widely used to effectively cope with the major challenges to the distribution network communication network. However, the distribution convergence terminal is at the edge of the network and the possibility of network attack is large and the traditional detection methods are mostly used after the occurrence of accidents, which cannot achieve timely detection and early warning of terminal abnormal behavior. In this paper, the non-intrusive distribution convergence terminal anomaly detection method is proposed. Specifically, the power consumption information of the main board is collected and analyzed in the device side channel. One normal program and one abnormal program are run respectively to extract the equipment status information hidden in the power consumption data. Moreover, in order to achieve more accurate equipment anomaly detection, the core algorithm based on LSTM neural network system is proposed to learn the data for a long time and build a prediction model. The simulation results show that the algorithm proposed in this paper has high accuracy and low error.