VFD-AE: Efficient Attack Detection in Industrial Cyber-Physical Systems using Vital Feature Discovery and Deep Learning Technique

The open environment of wireless communication network rapidly increases the attack surface of industrial cyber-physical systems (ICPS), making it vulnerable to various types of malicious attacks. Traditional intrusion detection systems are not applicable since in most cases they might not consider the lightweight treatment in feature space. However, the common feature selection algorithms often ignore the intrinsic relationships between features, which results in the failure of identifying irrelevant or redundant features. Motivated by this, we develop an efficient attack detection method using a complex network-based feature selection and the deep learning technique, referred to as VFD-AE. Specifically, we mine the potential correlations of features and realize effective feature selection by adopting the importance assessment technique in complex network theory. Furthermore, due to the attack sample short in supply often, an unsupervised detection scheme with Autoencoder (AE) is designed. With the proposed method, we can learn the attack mode efficiently and detect the attack effectively in ICPS. Simulation examples based on TE process (TEP) are carried out to verify the effectiveness of the proposed attack detection algorithm.