False data injection attacks detection in smart grids based on TCN prediction model
Weiyin Ma, Hong Wen, Wenjing Hou, Ruimin Yao, Yanxu Zhu, Dibao Yan, Ziang Zhao · 2024
The injection of false data attack (FDIA) poses a serious threat to smart grids, and accurate detection of FDIAs is crucial for the secure and stable operation of the grid. A promising approach for FDIA detection based on edge computing is to adopt a scheme that involves prediction before classification, taking into account the temporal correlations of measurement data. This paper proposes a FDIA detection model based on temporal convolutional network (TCN). The model effectively captures long-term dependencies in sequential data, thereby enhancing data prediction performance. Simulation results demonstrate that TCN exhibits superior detection performance under common neural network architectures, compared to typical recurrent neural network (RNN) models. This improvement facilitates highlighting the differences between injected false data and genuine data during the subsequent classification stage, making anomaly detection algorithms more capable of identifying abnormal situations and thereby enhancing the accuracy of FDIA detection.