l^1鲁棒辨识:最小二乘算法及试验设计

l^1 robust identification: least squares algorithm and experimental design

现有的l^1鲁棒辨识方法依赖于观测数据自的起始时刻因而不能用来辨识时变系统,针对该问题基于最小二乘法提出了一种l^1鲁棒辨识算法。该算法与观测窗的起始时刻无关,可用于时变系统的辨识,证明了当试验输入为持续激励信号时所提出的算法为本质最优算法,进一步证明了周期持续激励序列为最优试验信号,并给出了辨识误差紧界的计算公式。

Based on least squares algorithm a new l1 robust identifying approach was proposed, which was independent of the starting time of observation windows, and therefore, could be used to identify the time varying system. It was shown that the proposed algorithm was essentially optimal when the experimental input was selected as a persistent signal. Furthermore, the periodic persistent signal was proved to be the optimal experimental signal. Finally, the proposed algorithm was applied to slowly varying system.