Decentralized event-triggered reinforcement learning control for partially unknown nonlinear systems with time-varying states and asymmetric input constraints

This paper proposes a decentralized event-triggered control scheme based on reinforcement learning to stabilize a class of partially unknown nonlinear mismatched interconnected systems with time-varying state constraints and asymmetric control input constraints. First, a meticulously designed barrier function, which combines traditional and smoothing functions, transforms a constrained interconnected system into an unconstrained one. To address the impact of mismatched interconnections, auxiliary systems are designed for each subsystem. Additionally, non-quadratic utility functions are employed to constrain the inputs of these auxiliary systems. By solving the optimal control scheme for the auxiliary systems, a decentralized control scheme for the integrated mismatched interconnected system is achieved. Unlike traditional actor–critic network structures, the proposed identifier–critic network structure relaxes the constraints on system dynamics and eliminates errors caused by approximating the actor-network. The weight vectors in the critic network are updated using gradient descent and concurrent learning techniques, eliminating the need for traditional persistent excitation conditions. Furthermore, an event-triggered mechanism is introduced to reduce computational load and communication overhead. According to Lyapunov stability theory, it is rigorously proven that all signals of the interconnected nonlinear system are bounded. Finally, simulation examples validate the effectiveness of the proposed control scheme.