Neural-Network-Based Adaptive Optimized Finite-time Control of Switched Systems with Constraints

Abstract

This paper presents a finite-time optimized backstepping control strategy for uncertain strict-feedback switched systems subject to full-state constraints. The proposed method integrates adaptive backstepping, Nonlinear Mapping (NM), and reinforcement learning (RL) to achieve enhanced control performance. The NM technique is employed to prevent constraint violations and to relax the feasibility conditions. The RL framework adopts an identifier–actor–critic architecture, where the identifier estimates the unknown dynamics, the actor generates control inputs, and the critic evaluates system performance. To enhance computational efficiency, a simplified RL algorithm is developed, in which the update laws for the actor–critic network weights are derived from the negative gradient of a positive-definite function obtained through the partial derivative of the Hamilton–Jacobi–Bellman (HJB) equation. Simulation results demonstrate that the proposed control method achieves superior performance compared to state-of-the-art methods reported in recent literature.