Event-triggered prescribed finite-time optimal control for perturbed MIMO systems with dead-zone input via actor–critic reinforcement learning

This paper studies the prescribed finite-time optimal output feedback control issue for perturbed multiple-input multiple-output systems (MIMOs) via the actor–critic reinforcement learning technique. The main challenge faced is the robust design of an optimal event-triggered controller under the unknown dead-zone nonlinearity. To enhance the robustness of deterministic systems, a disturbance estimator is synthesized to offset the effect of external disturbances. On this foundation, since the suggested robust optimal scheme requires not only the training of the actor–critic learning laws but also the adaptive law of the disturbance estimation, both designing control algorithms and deducing learning laws potentially add to the sophistication of the control process. In addition, this paper sheds light on the solution for confining tracking error to a narrow feasible area, including overshoot and convergence time by resorting to asymmetrically parallel boundaries. Then, a robust prescribed finite-time optimal backstepping paradigm is developed, where the calculation explosion issue is circumvented with the aid of the second-order sliding mode integral filter. Notably, a switching event-triggered protocol integrating a parameter update law is formulated to execute the online compensation of the input dead zone and minmize needless resource wastage. Technically, the controller can ensure that the internal signals of closed-loop systems (CLSs) remain bounded, and the tracking error can be confined to an expected transient-state configuration before a finite time. Finally, two illustrative experiments exemplify the efficacy and superiority of the suggested control tactic. • SETM is constructed to balance resource usage and control behavior. • The prescribed finite-time function can restrict transient tracking error overshoots. • Disturbance estimation is embedded in actor-critic learning to boost the robustness.