非线性多输入系统的近似动态规划H_∞控制

There exist all kinds of multi-input systems in practical engineering, such as multi-driven servo system for large radar and artillery, multi-degree manipulator system and so on. Although the approximate dynamic programming (ADP) has been used to solve the various optimal control problems, it has not involved in the designation of the H_∞ controls for multi-input system. In this paper, the ADP method is used to design the H_∞ controllers of the multi-input nonlinear system. The neural network (NN) based on reinforcement learning is applied to study the solution of nonlinear Hamilton-Jacobi-Isaacs (HJI) equation. A new adaptive law is introduced to update the NN weights, which is then directly used to design the H_∞ controllers. The convergence of weights and the stability of the system are proved, which guarantees the good performance of the multi-input system under the unknown disturbance. Finally, a simulation example is given to illustrate the correctness and effectiveness of the proposed method.