RL‐Based H∞ Optimal Tracking Control for Markov Jump Singularly Perturbed Systems

ABSTRACT In this article, the reinforcement learning (RL)‐based optimal tracking control for a class of discrete‐time Markov jump singularly perturbed systems (MJSPSs) is investigated under the conditions of unknown system dynamics and external disturbances. First, an augmented system is constructed by integrating the reference model with the system dynamics, upon which a zero‐sum game performance index is established. Consequently, the corresponding coupled game algebraic Riccati equation (CGARE) and the ‐function Bellman equation are derived. Subsequently, a model‐free ‐learning algorithm is proposed to approximate the optimal solution of the CGARE by utilizing collected input‐state data, eliminating the reliance on precise system models. Finally, the effectiveness of the proposed method is validated through a tunnel diode circuit simulation, where stable tracking performance is achieved in the presence of external disturbances.