A Secure Transmission and Attack Detection Approach Based on Autoencoder

This paper is concerned with the secure transmission and attack detection problem for cyber-physical systems with unknown dynamic matrices. The signals transmitted between physical and cyber layers can be eavesdropped and injected with the false data by adversary, which causes the information leakage and performance degradation. To tackle this problem, an integrated secure transmission and attack detection framework is proposed. The system information is first hidden as latent variables for secure transmission without information loss by using the designed stable image representation-aided autoencoder. Moreover, to further enhance the security of transmission, the chaotic oscillators are also embedded as a supplement. Then, an attack detector is designed based on the residual generated from the original data and the reconstructed data. Compared with the existing results, the proposed method shows better encoding ability and detection performance. Furthermore, the attack types are further analyzed and identified by using an explainable artificial intelligence technique. Finally, two examples are given to show the effectiveness of the proposed defense method.