Gain-scheduling direct Heuristic Dynamic Programming, convergence analysis and application on Wind Turbine´s pitch control

This paper focuses on a nonlinear on-line learning control system with some priori knowledge that can be represented with the functions of the gain scheduling. For this, we develop the Gain-Scheduling direct Heuristic Dynamic Programming (GSHDP) controller that is based on the fundamental principle of reinforcement learning. In order to improve the convergence speed of the on-line learning, DL-RM algorithm is proposed in this paper. Moreover, we analyse the convergence characteristic of the algorithm and develop a more general convergence theorem. We apply the GSHDP controller on Wind Turbine´s pitch control and the performance of the controller is measured in our simulation experiments. The simulation results show that DL-RM algorithm has faster convergence speed than RM algorithm and proper selection of reinforcement signal can achieve better performance.