简练模型结构的辨识

Identification of Parsimonious Model Structure

中文关键词  辨识  模型结构  参数估计

英文关键词  identification  model structure  parameter estimation

中文摘要

      本文采用投影技术辨识系统的标准向量差分方程的简练结构---可观性指数，噪声自回归阶数以及需估的最小模型参数位置。本文提出的方法可在有色噪声下同时把可观性指数与AR模型噪声的阶估计出来。此外，算法还可确定那些对那些改善模型质量贡献最大的参数位置。

英文摘要

      The projection technique is used to identify a parsimonious model structure of canonical vector difference equations for an actual system. The structure consists of observability indices, autoregression orders and a parsimonious structure of parameters to be estimated. A new procedure, with a whitening projection operator, is proposed to estimate observability indices as well as autoregression orders in the case of coloured equation errors, which can be modelled with AR. Moreover, an algorithm is presented to determine the parsimonious parameters structure, each element of which has a significant contribution to improve the quality of the model.