不确定非线性系统的LSSVR迭代学习辨识算法  

Iterative Learning Identification Based on LSSVR for Uncertain Nonlinear Systems

作　　者：司利云[1,2] 余强[3] 林辉[2] 

机构地区：[1]长安大学电子与控制工程学院,西安710064 [2]西北工业大学陕西省微特电机及驱动技术重点实验室,西安710072 [3]长安大学汽车学院,西安710049

出　　处：《控制工程》2015年第6期1118-1122,共5页Control Engineering of China

基　　金：陕西省科技发展计划项目(2014K05-26);陕西省微特电机及驱动技术重点实验室开放基金(2013SSJ2003)

摘　　要：针对一类不确定非线性系统的时变参数辨识问题,利用最小二乘支持向量回归机(LSSVR)的逼近能力,将积分型Lyapunov函数作为优化目标,设计出具有自适应性的迭代学习辨识算法。该算法采用迭代学习的方式对支持向量回归机的权值和逼近误差进行修正,利用高增益的反馈控制器增强系统的稳定性和鲁棒性。收敛性分析表明,随着迭代次数的增加,估计误差渐近收敛于零。仿真结果表明该算法可以实现有限时间区间上参数的完全辨识。Aiming at the time-varying parameter identification problem for a class of uncertain nonlinear systems, by using least squares support vector regression （LSSVR） approximation ability, taking the integral Lyapunov function as the optimization objective, adaptive algorithm with iterative learning identification was designed. In the way of interactive learning, this algorithm modifies the weight and approximate errors of the support vector regress machine, and the system＇s stability and the robustness are strengthened by the feedback controller with high gain. The convergence analysis showed that, with increasing the number of iterations, the estimated error is asymptotic to zero. Simulation results showed that the algorithm can realize a complete identification of parameters within a finite interval.

关 键 词：迭代学习辨识 不确定非线性系统:在线建模 支持向量机 鲁棒控制 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]