Anomaly Detection on Time-series Logs for Industrial Network

With the deep integration of industrialization and informatization, the network environment is becoming more and more complex, and security is facing a huge threat. Recently, the industrial control systems pose an open trend, so the strategy of preventing external attacks through “physical isolation” does not work anymore. The security threats in the traditional IT field gradually affect the security of industrial control networks. Recently, more and more researchers apply artificial intelligence algorithms and blockchain technology to industrial control network security. This paper aims to propose a new way of thinking, starting from two levels of physical topology and time series structure for a specific industrial control system, establish a graph data structure, and then use the graph neural network (GNN) algorithm to detect abnormal nodes. We evaluate our approach through comprehensive experiments and the results are promising.