摘要
对于时／频混合数据的鲁棒辨识问题，提出了含有线性矩阵不等式的凸规划算法，并估计了系统的全局ｗ ｏｒｓｔ－ｃａｓｅ 误差界。仿真算例说明了该算法的有效性。
In system identification, it is important and practical to make full use of both time domain and frequency domain data. Ref.[2] provides an interpolation algorithm for stable linear shift invariant causal system based on mixed time/frequency domain data; but when the amount of data is large, it suffers from the following two shortcomings: (1) the computation is quite complicatied; (2) the order of identified model is usually high. We present a new algorithm to overcome these shortcomings. Our algorithm is based on convex programming. Eq.(14) is the target collection of identified models. Based on lemmas 1 and 2 [3] , we get theorem 1. From theorem 1, we can reduce the identification problem to a convex programming problem with LMI (Linear Matrix Inequality) constraints as shown in eq.(16), which is our main results. Solving eq.(16), we can get the identified model. Inequality (17) gives the explicit global worst case identification error bound in the H ∞ norm. Eq.(18) shows that our algorithm is convergent. The solid curves in Figs.1 and 2 are the responses of the real system in time domain and frequency domain respectively, while the dotted curves, those of the identified model. They show that the identified model is a good approximation of the real system.
出处
《西北工业大学学报》
EI
CAS
CSCD
北大核心
1999年第4期544-549,共6页
Journal of Northwestern Polytechnical University
基金
国家自然科学基金
关键词
鲁棒辨识
时/频混合数据
凸规划
系统辨识
robust identification, mixed time/frequency domain data, convex programming, worst case error
