Q‐Learning for Finite‐Horizon Decentralized Linear‐Quadratic Optimal Control Problem

ABSTRACT This article studies the finite‐horizon decentralized linear‐quadratic optimal control problem of large‐scale systems. The main feature of this article is that there are two controllers in the system, and each controller can only access its own measurement information, but not that of the other controller. In particular, the system matrices are unknown when designing the controller. The contribution of this article is designing a Q‐learning algorithm for the optimal control problem by deriving decentralized Q‐functions based on different measurement information. The effectiveness of the proposed algorithm is verified by two numerical examples.