A Non-normal Warning System for Dam Operation Using Machine Learning

Abstract

A country's critical infrastructures are heavily related to the quality of life and safety of the people. As a result, the security protection aspect of critical infrastructure has gained more and more attention nowadays, especially the security of its industrial control system (ICS). To avoid the abnormal condition happening in the critical infrastructure which could put people in great danger, a system that is capable of detecting any abnormal state of the ICS promptly is needed. Fortunately, due to the dramatic growth of the applications of machine learning in recent years, some researchers have already proposed anomaly detection methods with machine learning to provide instant warning and protection for ICS. However, most of the existing anomaly detection research tends to only target one cause that harms the system, such as attacks on the network or physical equipment failures. The ICS will be more comprehensively secured if the anomaly detection system can cover multiple aspects of the ICS. Therefore, we have established a non-normal warning system with the Generative Adversarial Network (GAN) for dam operations in this study, which can detect various types of non-normal operations and notify relevant personnel right away. Note that we use real historical data to make predictions and verify our warning system, and we improve it even more by implementing the visual analysis method, which makes up the indecipherable results often found in unsupervised learning.