Q-Learning Based Parameter Tuning for Model-Free Adaptive Control of Nonlinear Systems

In this paper, a Q-learning based parameter tuning method is proposed for the model-free adaptive control (MFAC) of nonlinear systems with unknown dynamics. Based on the partitioned state and action spaces, a novel reward function is firstly introduced to establish a Q-table offline training algorithm. Then, an online parameter optimization procedure is developed by means of the established Q-table, so as to enhance the learning ability and adaptability of the control system. Finally, an illustrative example is given to demonstrate the effectiveness of the proposed method in comparison with the existing parameter tuning methods of MFAC.