关键词: 规范变量分析, 子空间辨识, 奇异值分解, Akaike信息判据

Abstract: The research on modern control theory such as filtering, prediction and control mostly builds on the form of system state-space model. However, the traditional identification approach of system state-space model needs a priori parameterization and complex calculation. This paper made a deep research and extending on a subspace identification algorithm based on canonical variate analysis (CVA). In addition, a new Akaike’s information criterion is introduced to compute the state order, its computation is very simple and the result is comparatively exact. Since the state-space model of the system is generated directly from the input-output data, this algorithm avoids a priori parameterization of the state-space model. In respect of calculation, this algorithm relies mostly on the singular value decomposition, and therefore does not suffer from the numerical difficulties associated with the classical approach. The results of simulation prove that this method can very precisely identify the model of the system after using the new approach of compute state order, so it has a quite broad foreground.

Key words: Canonical Variate Analysis, subspace identification, singular value decomposition, AIC