摘要
为了克服传统扩展卡尔曼滤波算法进行参数估计时可能产生的新数据失效问题,提出了一种改进的扩展卡尔曼滤波(EKF)步骤,然后将改进步骤做为人工神经网络的学习算法用于基于前向神经网络的非线性时变系统辨识。与传统的扩展卡尔曼滤波步骤相比克服了新数据的饱和现象,可以更好地反映系统时变特征。通过一个单变量一般时变非线性系统和一个三自由度非线性时变刚度结构系统算例,仿真验证了新算法在辨识精度和计算量方面的改进特性。
In order to overcome the drawback of traditional extended kalman filter (EKF) procedure,the improved EKF scheme was proposed,and applied to nonlinear time-varying system identification based on feed forward neural network.The proposed algorithm relaxes the data saturation.Simulation results of both a nonlinear time-varying systems with a single variable and a three degrees-of-freedom structural system with nonlinear time-varying stiffness show that the proposed algorithms can overcome the problem of divergence and consume less computational cost and is of higher accuracy and robustness.
出处
《振动与冲击》
EI
CSCD
北大核心
2010年第8期5-8,共4页
Journal of Vibration and Shock
基金
国家自然科学基金(10672045)
新世纪优秀人才支持计划(NCET-06-0344)
关键词
非线性时变系统
多层前向神经网络
系统辨识
改进扩展卡尔曼滤波算法
nonlinear time-varying system
multi-layer feed forward neural network
system identification
improved extended kalman filter EKF
