Reinforcement Learning for Mean-Field Linear Quadratic Control with Partial System Information
Yingxia Lin, Qingyuan Qi · 2023
In this paper, the infinite horizon linear quadratic (LQ) optimal control problem for mean-field systems with partial system information is solved by using the reinforcement learning (RL) approach. Although the introduction of the mean-field terms in system dynamics and the cost function will destroy the adaptiveness of the control law, the optimal stabilization control is derived based on the proposed online RL algorithm and the Bellman dynamic programming. Moreover, the proposed algorithm requires only the local state data of the mean-field system to compute the optimal control, and the stepwise stability of the stabilizers is shown. Finally, numerical example is given to illustrate the effectiveness of the proposed algorithm.