Reinforcement learning-based prescribed-time optimized adaptive fuzzy control for multi-agent systems with output saturation

This paper investigates the self-triggered prescribed-time optimized adaptive fuzzy control issue for multi-agent systems (MASs) with output saturation. The main challenge is to design an adaptive optimal control algorithm that can mitigate the issue of unknown control gains generated by output saturation. First, the non-smooth output saturated nonlinearity is captured by a smooth model that counts on the continuity of the hyperbolic tangent function, allowing the output signal to be available for the recursive control process. Additionally, the uncertain gain can be skillfully circumvented by building the Hamilton-Jacobi-Bellman (HJB) equation oriented towards the state signal rather than the error variables to tackle the conflicting issues. Then, the reinforcement learning-based (RL) prescribed-time control strategy is delicately engineered, thereby avoiding the calculation sophistication and excluding the drawback of semi-global boundedness of the error surface with the assistance of the designed prescribed-time filter and Lyapunov-like energy candidate. Notably, a peculiar self-triggered control protocol considering auxiliary functions is synthesized to conserve resource loss and stay away from the sustainable monitor criteria and the singularity dilemma. Benefiting from the prescribed-time bounded stability theory, it can be guaranteed that the closed-loop systems (CLSs) realize the prescribed-time stability, and the system outputs are restricted in their saturated thresholds. Herein, the suitability and rationality of the investigated tactic are exemplified through simulation instances.