Discrete-Time Reinforcement Learning Adaptive Control for Non-Gaussian Distribution of Sampling Intervals.

Abstract

This article proposes an optimal controller based on reinforcement learning (RL) for a class of unknown discrete-time systems with non-Gaussian distribution of sampling intervals. The critic and actor networks are implemented using the MiFRENc and MiFRENa architectures, respectively. The learning algorithm is developed with learning rates determined through convergence analysis of internal signals and tracking errors. Experimental systems with a comparative controller are conducted to validate the proposed scheme, and comparative results show superior performance for non-Gaussian distributions, with weight transfer for the critic network omitted. Additionally, the proposed learning laws, using the estimated co-state, significantly improve dead-zone compensation and nonlinear variation.