Composite resilient control for the USV system against unknown multiplicative sensor FDI attacks: An adaptive deep reinforcement learning approach

In this article, the composite resilient heading tracking control problem is investigated for the unmanned surface vehicle (USV) under unknown multiplicative sensor false data injection (FDI) attacks. First, a resilient compensation strategy is proposed to counter these malicious behaviours. Specifically, an adaptive compensator is designed to offset the attack signals and reconstruct the corrupted tracking error variables, thereby effectively mitigating the adverse impacts of FDI attacks. Second, an adaptive deep reinforcement learning (DRL) framework is developed, which embeds online adaptive deep neural networks into an RL-based optimal control structure. This embedded architecture not only improves the approximation performance for the uncertain USV dynamics but also considerably enhances the system’s adaptability in complex environments. Finally, the proposed scheme ensures precise and secure heading tracking performance, guaranteeing that all closed-loop signals are uniformly ultimately bounded (UUB) via Lyapunov stability theory. Simulations and experiments validate the effectiveness of the proposed approach.