Long Short-Term Memory Network-Based H∞ Synchronization Control and Anomaly Detection for Cyber-Physical Systems

Abstract

In the synchronization of cyber-physical systems (CPSs), modeling the nonlinear dynamics of physical plants is a challenging task. To address this challenge, we propose a novel H∞ controller design method that leverages a data-driven approach to robustly synchronize CPSs and ensure their stability. In the proposed approach, the input-output relationship of the physical system is learned using long short-term memory (LSTM) networks to approximate the unknown dynamics of CPSs. Furthermore, we exploit an effective control scheme for trained LSTM networks to effectively handle the nonlinearity of activation functions. To ensure stability and performance in the convergence of synchronization error, a controller design criterion is derived for the trained LSTM network in terms of linear matrix inequalities, and the controller gain is computed using convex optimization techniques. In addition, we present an anomaly detection algorithm using the proposed method, which can synchronize CPSs and detect abnormal signals without requiring any prior physical model information. Consequently, the stability of the synchronization control system can be ensured, enabling its application to anomaly detection. Finally, the effectiveness of the proposed method is validated through an experiment on a motor control system even in abnormal operating conditions.