基于梯度估计的非线性系统最优控制及仿真①-计算机系统应用.pdf

计 算 机 系 统 应 用 2016 年 第 25 卷 第 11 期 ① 基于梯度估计的非线性系统最优控制及仿真 缪应锋, 姚庆华, 李智雄, 宋晓轩 (云南省交通科学研究院, 昆明 650093) 摘 要: 基于自适应动态规划(ADP)执行-评价结构, 应用神经网络(NN)对非线性系统进行最优控制求解. 首先提 出所求解非线性系统的一般形式; 其次给定二次正定性能指标, 求其哈密尔顿函(HJB)函数; 分别应用神经网络 对执行-评价结构中的性能指标和最优控制进行逼近, 神经网络权重参数应用梯度法求得, 从而可以求得其最有 控制策略. 而且对执行机构和评价机构神经网络权重参数的收敛性以及系统总体的稳定性进行了详细的分析, 证明所求控制策略可以使系统稳定; 最后, 用仿真结果来验证所提出的方法的可行性. 关键词: 非线性系统; 动态规划; 梯度算法; 哈密尔顿函数; 李雅普诺夫 Optimal Control and Simulation of Nonlinear Systems Based on Gradient Estimation MIAO Ying-Feng, YAO Qing-Hua, LI Zhi-Xiong, SONH Xiao-Xuan (Institute of Yunnan Traffic, Kunming 650093, China) Abstract: This paper solves the regulation optimal control problem for nonlinear systems based on the adaptive dynamic programinf(ADP) and neural networks(NNs). Firstly, we propose the regulation nonlinear system; afterwards, the cost function and its Hamilton-Jacobi-Bellman(HJB) function were given; furthermore, an actor-critic frame structure was put forward to get the optimal control, and the neural network algorithms were utilized to approximate the performance index and the optimal control ofactor-critic structures respectively. The NN weights are estimated by gradient algorithms, so that the optimal control is obtained. Simultaneously, the stability of the whole system and the convergences of actor-critic NN weights are proved based on the Lyapunov theory. Finally, the simulation results are provided to verify the effectiveness of the proposed methods. Key words: nonlinear system; approximate dynamic program (ADP); gradient algrothim; Lyapunov; Hamilton-Jacobi -Bellman(HJB) 最优控制求解分为线性系统和非线性系统. 线性 致计算参数收敛速度比较慢, 系统稳定需要很长时间. 系统用里卡提方程求解. 对于非线性系统,