アイテム詳細

要旨

This article studies the optimized adaptive finite-time consensus control issue for stochastic nonlinear multiagent systems subject to non-affine nonlinear faults. Under the architecture of the adaptive optimized backstepping method, this article develops the neural-network-based simplified reinforcement learning algorithm with an identifier-critic-actor structure, where the identifier, critic and actor are put forward to estimate unknown dynamics, evaluate system performance and implement control behavior, respectively. Then, the Butterworth low-pass filter is introduced to compensate for the adverse effects brought by non-affine nonlinear faults. Furthermore, it is verified by Itô differential equation and the finite-time theory that the closed-loop system is semi-global finite-time stable in probability. Finally, the effectiveness of the control algorithm is illustrated by simulation examples. Note to Practitioners —This paper was motivated by the problem of finite-time convergence is one of significance performance index in many practical application. For systems with high transient performance standards, such as robotic systems, manipulator systems and unmanned aerial systems, finite time convergence is of practical importance. Accordingly, distinguished from the previous investigation results, this article develops the neural-network (NN)-based simplified reinforcement learning (RL) algorithm with an identifier-critic-actor structure, where the identifier, critic and actor are put forward to estimate unknown dynamics, evaluate system performance and implement control behavior, respectively. We believe that the novel research method will bring a research spring for the constrained systems. Preliminary simulation experiments suggest that this approach is feasible. In future research, we will address the fixed time control protocol designs for nonlinear multi-agent systems.