基于前向神经网络的时变非线性结构系统辨识快速递推最小二乘法  

Fast recursive least square scheme for time-varying nonlinear structural system identification based on feed forward neural network

作　　者：于开平[1] 董好志[2] 

机构地区：[1]哈尔滨工业大学航天学院,黑龙江哈尔滨150001 [2]中国电子科技集团公司第38研究所,安徽合肥230031

出　　处：《振动工程学报》2007年第5期468-472,共5页Journal of Vibration Engineering

基　　金：国家自然科学基金资助项目(10672045);新世纪优秀人才支持计划(NCET-06-0344)

摘　　要：针对基于前向神经网络的普通递推最小二乘估计存在着自适应跟踪慢和精度低的问题,提出了一种可对非线性时变系统进行快速辨识的新方法,因该方法有类似递推最小二乘算法的形式,称其为基于前向神经网络的快速递推最小二乘算法。该算法对传统的递推最小二乘算法的递推方式进行了改变,以更好的跟踪非线性时变系统的动态特性。针对典型的系统辨识仿真算例,通过与现有常用方法的比较研究显示了这种算法具有计算简单、收敛速度快和辨识精度高的良好性能。最后将方法用于一个三自由度时变非线性振动系统,结果同样验证了方法的良好特性。In this paper the new recursive least square scheme for time-varying nonlinear structural system identification based on feed forward neural network model is presented. The new scheme improves adaptive tracking property by modifying recursive factor in the recursive least square procedure. The performance of the algorithm is evaluated on the identification of some typical time-varying nonlinear system including the three degrees of freedom structural system with nonlinear time-varying stiffness. Simulation results demonstrate that the proposed technique is capable of fast tracking the system parametric change and reducing the computational cost largely.

关 键 词：非线性时变系统 系统辨识 快速递推最小二乘算法 神经网络 结构系统 

分 类 号：O322[理学—一般力学与力学基础]