基于神经网络近似的自适应优化控制

Adaptive Optimal Control Based on Neural Network Approximation

为了解决初始和终端确定的一类离散时间非线性系统有限时间优化控制,利用动态规划原理求解过程中遇到维数灾的问题,提出了基于神经网络的自适应动态规划近似优化控制。在分析动态规划求解遇到维数灾的基础上,进而给出了迭代ADP算法,并采用神经网络近似代价函数和控制律来实现迭代ADP算法,设计近似优化控制器。通过mat lab实验仿真结果表明,采用迭代ADP算法能够避免求解中遇到的维数灾,从而有效地实现了一类离散时间非线性系统的有限时间近似优化控制。

In order to solve the finite optimal control for a class of discrete-time nonlinear system with the initial and terminal certain, meeting the problem of dimension disaster in solving by dynamic programming,an adaptive iterative approximate control method based on neural network is proposed. Analyze the problem of the dynamic programming solving meets dimension disaster,then give the iterative ADP algorithms, which is realized by neural network approximate cost function and control law, by the iterative ADP algorithm design the approximate optimal controller. Through the matlab experimental simulation, the results show that the iteration ADP algorithm can avoid the problem of the dimension disaster, effectively realized the finite approximation optimal control for a class of discrete-time nonlinear system.