Journal
INFORMATION SCIENCES
Volume 484, Issue -, Pages 84-94Publisher
ELSEVIER SCIENCE INC
DOI: 10.1016/j.ins.2019.01.060
Keywords
Simultaneous policy update algorithm; H-infinity control; Continuous-time algebra Riccati equation; Kantorovich's theorem; Reinforcement learning
Categories
Funding
- National Natural Science Foundation for Distinguished Young Scholars of China [61625302]
- National Natural Science Foundation of China [61473011, 61721091]
Ask authors/readers for more resources
In this paper, we present a new insight into the simultaneous policy update algorithm (SPUA) for solving H-infinity control problems. From the perspective of practicability, a checkable criterion is presented for the initialization of SPUAs such that the initial guess can be chosen with clear rules and lie within a local domain of convergence. To further get a better convergence property, a novel initialization is developed such that the starting matrix does not need to lie in a neighborhood of the solution of H-infinity control problems in some cases and its convergence is established rigorously by mathematical induction principle. Subsequently, an improved SPUA is proposed, and its model-free variant based on reinforcement learning (RL) is also able to learn the solution online without any system dynamics. Finally, numerical results illustrate the effectiveness of the proposed methods. (C) 2019 Elsevier Inc. All rights reserved.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available