Data-driven multi-model predictive control for nonlinear systems under cyber attacks

Model predictive control has demonstrated significant potential in managing nonlinear systems, but its effectiveness remains vulnerable to sophisticated cyber attacks. This paper presents a novel data-driven multi-model predictive control (DMMPC) framework that synergistically integrates cyber attack resilience with temporal feature learning. Compared with existing methods that focus on isolated channel attacks, the proposed framework explicitly considers cross-channel interference effects, enabling simultaneous mitigation of cyber attacks in sensor-controller and controller–actuator channels. Firstly, a data-driven anomaly detection system combining historical pattern matching with real-time signal deviation analysis is proposed to decrease the effects of sophisticated cyber attacks. Then, an expectation-based DMMPC method for nonlinear systems is designed to address the cyber attacks, and the bounded-input bounded-output stability of the closed-loop system is theoretically proven. Finally, the effectiveness of the proposed method is validated through numerical simulations and mobile robot experiments. Experimental results show that the proposed framework maintains tracking accuracy and system stability under various attack scenarios.