Stacked-Autoencoder Based Anomaly Detection with Industrial Control System

摘要

The Industrial Control System (ICS) is a system for controlling industrial systems. It is mainly a national infrastructure, and if it is shut down, it can have a huge impact on our lives. Therefore, ICS is mainly operated in a closed network to minimize security threats. However, ICS has also increased its Internet connection points as the IoT advances, which has increased security threats. Until now, it was difficult to secure a data set from an actual operating environment in ICS, so it was difficult to study effective security techniques. In this paper, we proposed a stacked-autoencoder (SAE), deep Support Vector Data Description (SVDD)-based data anomaly detection technique using an ICS dataset created based on a testbed similar to an actual operating environment, and derived detection accuracy for each threshold. In both models, the highest accuracy was derived when the threshold was 0.98, and the accuracy was 96.03% in the SAE model and 95.48% in the Deep SVDD model.