The optimal control problem with nonlinear disturbance is abstracted by Wind Turbine pitch control problem. The optimal control problem has the stochastic disturbance with complex nonlinear, so there are some difficulties when ADP (Approximate Dynamic Programming) method is used to solve the problem. On the one hand, it is impossible to use on-line Policy Iteration method for feedback control of deterministic problem to solve the problem with nonlinear disturbance. On the other hand, when Action-Dependent Heuristic Dynamic Programming (ADHDP) method for the stochastic Markov Decision Process (MDP) problem is used in the problem with nonlinear disturbance, the on-line algorithm cannot converge.Considering the problems above, ADP method with nonlinear disturbance conpensation is researched to solve the optimal control problem with nonlinear disturbance in this thesis. The main works are described as below.(1) The Policy Iteration controller with feedforward disturbance conpensation is designed. The optimal control problem with nonlinear disturbance is divided into deterministic optimal control problem and disturbance conpensation problem. Policy Iteration controller is designed for the deterministic optimal control problem. Concretely, Linear ADP On-line Policy Iteration control with feedforward disturbance conpensation is designed for Linear Quadratic Regulator (LQR) problem with unknown model parameters and Nonlinear ADP On-line Policy Iteration control with feedforward disturbance conpensation is designed for nonlinear control problem with unknown model parameters.(2) GS-ADHDP (Gain-Scheduling Heuristic Dynamic Programming) control method is used for unknown model structure system with stochastic disturbance. The method is based on disturbance conpensation technology. In GS-ADHDP, GS functions whose independent variable is disturbance is embedded in action network of ADHDP as a-priori knowledge.. Under some certain conditions, convergence of stochastic multiple networks on-line learning system that is generalized from GS-ADHDP is analyzed.(3) WECS (Wind Energy Conversion System) model of Wind Turbine based on the blade element - momentum aerodynamic theory and pitch optimal control problem model with tubulence disturbance are established. After the control problem is analyzed and simplified, Linear ADP On-line Policy Iteration control with feedforward disturbance conpensation, Non-linear ADP On-line Policy Iteration control with feedforward disturbance conpensation and GS-ADHDP control are used in Wind Turbine pitch optimal control problem.(4) The simulation experiment and comparison test of linear off-line algorithm and ADP algorithms for Wind Turbine pitch control are simulated in MATLAB software. According to the comparison test results, the on-line ADP control is better than linear off-line control in the meaning of achieving control objective; non-linear ADP control method is better than linear ADP control method to achieve control objective and reduce Wind Turbine loads; Nonlinear ADP On-line Policy Iteration control is better than the GS-ADHDP control to achieve control objective. |