【摘要】 本文研究一个带时滞的动态多变量CARMA模型的结构及参数辨识。将一个具有R个输入,M个输出的多变量主模型,分解为M个带有R个输入的子模型,而每个子模型又可分解成R个子子模型。子模型与子子模型分别进行辨识,首先,用CAR模型输出逼近CARMA模型输出,以获取白噪声残差序列。然后,辨识CARMA模型。本文采用分块矩阵求逆公式,提出一种快速的依阶次递推的增广最小二乘法,并根据阶的判定准则和节省原理,可同时目动确定模型的阶次、时滞和节省参数。本算法比其它算法节省时间。更多还原

【Abstract】 The structure and parameters identification for a multivariablc dynamic CARMA model with deadtime is discussed. A main model with R inputs and M outputs is decomposed into M submodels. And then a submodei with R inputs is decomposed into R subsubmodels. The submodel and subsubmodel are identified separately. First, a CAR model output approaches to CARMA model output, then can obtain the while noise sequence. Finally, CARMA model is identified. This paper presents an algorithm that allows us to recursively compute the extended least-squares estimates (ELSE) using the property of partitioned matrix inverse as the number of parameters increases. Thus, according to the criterion of an order determination and the parsimony principle, the order, deadtime and parsimonious parameters of models can be estimated simultaneously and automatically. It takes much less computer time than other algorithms.更多还原