Enhanced Cyber-Physical Security through Deep Learning Techniques

Nowadays that various aspects of our lives depend on complex cyber-physical systems, automated anomaly detection, as well as attack prevention and reaction have become of paramount importance and directly affect our security and ultimately our quality of life. Recent catastrophic events have demonstrated that manual, human-based management of anomalies in complex systems is not efficient enough, underlying the importance of automatic detection and intelligent response as the recommended approach of defence. We proposed an anomaly detection framework for complex systems based on monitored data storage and Statistical Correlation Analysis for different pairs of constituent time-series of a multivariate time series segment, and unsupervised deep learn-ing to intelligently distinguish between normal and abnormal behavior of the system. Experimental results demonstrate that the proposed model is much better than baseline methods, and it can model (inter)correlation and temporal patterns of multivariate time series effectively.