Reinforcement Learning-Based Predefined-Time Adaptive Optimized Control for a Two-DOF Helicopter via Event-Triggered Communication

This paper focuses on reinforcement learning (RL)-based event-triggered optimized adaptive predefined-time control issue for a two-degree of freedom helicopter system. In the procedure of recursive design, an enhanced predefined-time nonlinear disturbance observer is constructed, which not only enhances the disturbance rejection ability of control system, but also ensures the convergence property of disturbance estimation error. Then, a RL-based optimization strategy with actor-critic-identifier (ACI) structure is incorporated into the predefined-time dynamic surface control framework, which guarantees that approximate optimal control can be achieved. In addition, the communication burden is significantly reduced by using a switching event-triggered mechanism. By utilizing the developed controller, it has been confirmed that all signals in the closed-loop system are practically predefined-time bounded. Finally, the availability and superiority are demonstrated by simulation results by the proposed control method.