AIGCN: Attack Intention Detection for Power System Using Graph Convolutional Networks

Power systems have been attracting the attention of attackers because of its great value. Identifying attack intentions is essential for proactively blocking the intrusion into power information systems. In this paper, we propose AIGCN, a novel attack intention detection model based on graph convolutional networks. Particularly, AIGCN first presents an abnormal IP detection method based on log behavior analysis to filter suspicious malicious IPs. And then AIGCN models the interactive relationships between suspicious IPs as a graph and performs graph convolution operation on the graph to effectively detect the attack intentions and learn the attack patterns with different intentions. Experimental results on real-world datasets verify that AIGCN outperforms baseline methods in detecting attack intentions and demystifying corresponding attack patterns.