Finite-Time Event-Triggered Optimal Tracking Control for Strict-Feedback Nonlinear Systems With Input Delay via Reinforcement Learning
TL;DR
An online optimization algorithm is developed to achieve practical finite-time optimal tracking, where a finite-time actor–critic law is introduced to inject the same error-dependent evaluative term into both networks, rendering the weight misalignment dynamics finite-time stable without persistence of excitation and accelerating neural weight convergence.
Abstract
This paper investigates the practical finite-time event-triggered optimal tracking control problem for strict-feedback nonlinear systems with input delay. By integrating an adding-a-power-integrator backstepping design with a reinforcement learning actor–critic framework, an online optimization algorithm is developed to achieve practical finite-time optimal tracking, where a finite-time actor–critic law is introduced to inject the same error-dependent evaluative term into both networks, rendering the weight misalignment dynamics finite-time stable without persistence of excitation and accelerating neural weight convergence. Furthermore, an event-triggered mechanism tailored to the input-delay setting is constructed, in which a Padé approximation compensates the delay and the control signal is updated only when a tracking-error-dependent threshold is violated. Lyapunov analysis verifies that the closed-loop event-triggered system is semi-globally practically finite-time stable and free of Zeno behavior. Comparative simulations on a permanent magnet synchronous motor drive confirm that the proposed scheme achieves high tracking accuracy, significant communication savings, and near-optimal control performance.