子空间系统辨识方法的系统阶数估计

摘要

阶数估计是子空间系统辨识方法的一个重要环节,一直以来没有得到很好解决.回顾了以奇异值序列"间断"点、奇异值梯度序列"间断"点、NIC准则和奇异值准则 (SVC)为判断依据的4种阶数估计方法.并在SVC的基础上提出改进方法MSVC.通过Monte Carlo试验比较了各准则的性能,结果表明MSVC的估计效果优于其它4种准则.并在试验结果的基础上给出了数据块Hankel矩阵的块行数和块列数的取值建议.%Four different estimation criterions, namely, the gap of singular value sequences, the gradient gap of singular value sequences, NIC and singular value criterion (SVC), were reviewed and an improved criterion MSVC was proposed based on SVC. The performances of these five criterions were compared through Monte Carlo test. The results indicate that the effect of MSVC is superior to that of others. Some suggestions were presented for the choice of the numbers of block rows and block columns of the data block Hankel matrix based on the test results.