Adaptive Event-Driven Model-Free Neural Learning Control for Unknown Nonlinear Systems With Dynamic Compensation

With the proliferation of networked systems in Internet of Things (IoT) infrastructures, efficient control strategies for information-constrained environments have become increasingly critical. This article investigates the event-driven model-free optimal control problem for completely unknown nonlinear systems by means of integral reinforcement learning (IRL) and compensation mechanism. For the original unavailable dynamics, a general-form input compensator is designed to establish a partially-unknown system model. To conserve limited communication bandwidth, a dynamic event-triggered scheme (DETS) is implemented, where only state data satisfying the predefined condition are transmitted over the network. Subsequently, a single critic structure-based IRL algorithm is proposed to address the Hamilton–Jacobi–Bellman equation (HJBE) with DETS, where the critic weights are tuned by an improved gradient descent method. By incorporating an adjustable state-dependent term into the update rule, the convergence of the critic weights is ensured without an initial admissible policy. Moreover, the experience replay technique is employed to release the persistence of excitation condition. The stability analysis of the reconstructed system is conducted in accordance with the Lyapunov principle. Eventually, the feasibility and practicality of the developed control scheme are validated through a simulation example.