Traffic Anomaly Detection for Smart Grid based on Hierarchical Spatial-Temporal Feature Learning
Guoli Feng, Ning Wang, Xinnan Ha, Xiaobo Li, Run Ma, Peng Lin · 2023
With the development of computer technology and communication technology, the demand for broadband smart grid services is rising rapidly, and the network traffic is growing rapidly. With the digital evolution of power grid, smart grid has also emerged, which makes broadband power network more widely used. In recent years, the traffic and bandwidth of smart grid have increased dramatically, smart grid is facing increasing risk of attack, traffic anomaly detection for smart grid is of great significance to the healthy development of broadband smart grid. This paper proposes a traffic anomaly detection method based on hierarchical spatial-temporal feature learning for smart grid service, using Res2Net and Spatial Attention Mechanism to improve CNN to extract spatial features, and then connect Bidirectional LSTM module to extract temporal features. This method extracts features from temporal granularity and spatial granularity. By adding multi-scale feature extraction module and spatial attention module, while obtaining spatial feature information of different scales, different weights are given to spatial features, and more attention is paid to the parts containing more key information, the part that contains less key information will be less concerned, thereby enhancing the ability of feature representation, improving the accuracy of traffic anomaly detection, and reducing the false alarm rate of smart grid.