用于线性时变系统辨识的固定长度平移窗投影估计递推子空间方法

给出了一个新的用于线性时变参数结构系统模态参数识别的基于固定长度平移窗投影估计的递推子空间方法。首先推得基于平移窗投影估计的子空间跟踪算法，以代替奇异值分解，再推得系统数据矩阵的一阶修正形式，从而得到新的基于平移窗投影估计的递推子空间方法。该方法可有效地降低算法的计算量。最后通过刚度随时间变化的3自由度系统和一具有移动质量的机械臂系统的时变模态频率辨识仿真表明该方法可有效地辨识线性时变系统的伪模态参数。

USING RECURSIVE SUBSPACE METHOD BASED ON PROJECTION APPROXIMATION WITH MOVING WINDOW TO ESTIMATE PARAMETERS OF LINEAR TIME-VARYING STRUCRURE

A novel recursive subspace method is developed based on fixed length moving window (FLMW) projection approximation used for estimating the modal parameter of linear time-varying structural system. The projection approximation subspace tracking (PAST) method with FLMW is developed to replace a singular value decomposition in order to decrease the computational complexity. A 3 degree-of-freedom model with time-varying stiffness and a manipulator with moving mass are used to testify these two methods. The simulation results show that these two methods can avoid the influence of the old data efficiently, and the novel recursive method can estimate the parameter of linear time-varying system as efficiently as the recursive subspace method with FLMW, with the computational complexity decreasing at the same time.