基于神经网络的最优保代价控制器设计  

Optimal Guaranteed Cost Controller Based on Neural Network Model

作　　者：刘飞[1] 周洪亮[2] 

机构地区：[1]江南大学自动化研究所,江苏无锡214036 [2]浙江大学控制系,浙江杭州310027

出　　处：《控制工程》2003年第5期391-394,共4页Control Engineering of China

摘　　要：针对一类非线性系统,并以典型倒立摆装置为对象,讨论了以一种稳定性分析为基础的控制器设计方法。首先利用多层神经网络模型逼近动态系统中的非线性环节,并表示成线性微分组合形式。在此基础上,结合线性系统二次稳定理论和线性矩阵不等式技术,引入线性二次型指标,给出保代价反馈控制器设计方法,进而优化代价指标的上界,可得最优保代价控制器。该设计方法中,神经网络建模和反馈增益求解均可利用现有的算法,简单易行,仿真结果表明了其有效性和适用性。Stability-based control design for a class of nonlinear dynamical systems is introduced. The approach is illustrated with a typical inverted pendulum. Firstly, nonlinearity of dynamical systems is approached by a multi-layer neural network, and then is treated as linear differential inclusions (LDI). Based on robust quadratic stability theory and linear matrix inequality (LMI) technique, a guaranteed cost index controller is presented. Furthermore, the optimal guaranteed cost controller is obtained while minimizing the cost bound. By means of common back propagation algorithm (BP) for net training and numeric solver for LMI, the results of simulation from this approach show the applicability and validiry.

关 键 词：最优保代价控制器 设计 神经网络 闭环系统 反馈增益 倒立摆装置 自动控制 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]