時空間情報に基づいた閉ループ系の直接同定法

抄録

A new identification algorithm is developed for direct closed-loop identification by using the temporal-spatial information extracted from the over-sampled input and output data. It has been shown that the poor convergence and numerical problems often arise in the general prediction error method (PEM) when the external exciting signals have not sufficient frequency components, or the input and output signals are strongly correlated with the noise. As a result, the closed-loop identification may fail to offer a valid plant model. In this paper, the distinct subspace characteristics of the sampled data in the output over-sampling scheme are analyzed, and the new criterion is presented for the numerical optimization. It illustrates that the convergence region is enlarged, while the condition number of data matrix is reduced in the new algorithm, so the performance of the identification algorithm can be improved, and the variance of the estimated parameters is decreased greatly. The simulation examples demonstrate the effectiveness of the proposed algorithm, and show that it has better performance than the conventional closed-loop methods especially under the severe numerical conditions.