A nonlinear optimal feedback controller using neural networks

A general optimal feedback controller can be obtained by solving the Hamilton Jacobi Bellman dynamic programming equation. But for a nonlinear dynamic system, it is a difficult task. We propose a practical and effective method for constructing an approximate optimal feedback controller, where multilayer neural networks are employed in identification of nonlinear systems and Taylor expansions are exploited to get the approximate optimal solution for the nonlinear feedback controller. Two examples are given to demonstrate the effectiveness of the proposed method.