Reinforcement Learning-Based Robust Tracking Control for Unknown Markov Jump Systems and its Application
Hao Shen, Jiacheng Wu, Yun Wang, Jing Wang
Abstract
This brief presents a novel reinforcement learning-based robust tracking control method for discrete-time unknown Markov jump systems. First, the optimal tracking and robust controller design problem is formulated as an optimal output regulation problem. In particular, we reconstruct the stochastic coupled algebraic Riccati equation to decouple the jumping mode and approximate the optimal control policy, where the knowledge of system dynamics should be known as a priori. To solve this problem, by employing the online reinforcement learning approach, the optimal output regulator is learned within a novel data-based parallel learning framework. On this basis, the solutions of the stochastic coupled algebraic Riccati equation and the output regulation equation of Markov jump systems are obtained by using online system data. Moreover, the convergence of the proposed algorithms is analyzed. Finally, a PWM-driven DC-DC boost converter model is provided to show the effectiveness of the proposed method and the main theoretical results.