Fuzzy-Boosted Event-Triggered Tracking Control of Unknown Nonlinear Networked Systems: A PSO-Driven RL Approach

摘要

This article is centered on an optimal tracking control problem for nonlinear networked systems with limited network bandwidth and unavailable dynamics. Foremost, considering the absence of system dynamics, a state identifier grounded in generalized fuzzy hyperbolic model (GFHM) is estabilshed to eliminate the reliance on known system states. Meanwhile, in order to efficiently utilize the limited communication resources within the network, a dynamic event-triggered mechanism (DETM) with an adaptive threshold is adopted. Subsequently, the DETM-based Hamilton-Jacobi-Isaacs (HJI) equation is constructed following the Bellman optimality theory. Furthermore, we develop a particle swarm optimization (PSO)-driven reinforcement learning (RL) algorithm to achieve the desired optimal tracking control strategy. It is demonstrated that the tracking error system can be promised to be uniformly ultimately bounded (UUB) stability. Eventually, a numerical experiment is provided to illustrate the validity of the developed RL algorithm.Note to Practitioners—This study targets the optimal tracking control issue of unknown networked systems, which holds practical significance in various fields such as robotics, autonomous vehicles and industrial control. In real-world environments, the dynamics of networked systems are not always accessible, thus this paper employs GFHM to approximate unknown dynamic information. Additionally, due to the involvement of the network, the restriction of communication resources is an unavoidable concern. To address this, DETM is adopted to improve network transmission efficiency. Subsequently, to achieve the optimal tracking control, the HJI equation based on DETM is constructed. The solution to the HJI equation is obtained through the RL algorithm supported by critic neural network, with its weights trained using the PSO algorithm. Compared to the commonly applied gradient descent method, under the PSO algorithm, the initial weights can be randomly assigned, making implementation easier and increasing the success rate of system operation. This is of significant importance in practical applications. Besides, a simulation example illustrates the feasibility of the developed method.