动态数据系统(DDS)建模方法的改进

本文概述了原DDS建模方法的缺点,叙述了改进的途径,推导了改进的算法,给出了改进部分的详细框图。首先用长自回归拟合数据直到残差为白噪声,如AR模型阶不高,则认定为AR模型。否则,利用数据和残差,线性估计ARMA模型参数。增阶和降阶时,Yule—Wolker方程系数阵和模型参数都可递推求得。初估ARMA模型后,用非线性最小二乘法来优化参数。通过二个实例计算表明,经改进的方法可以节省66％的机时,效果是明显的。

AN IMPROVED ALGORITHM FOR DYNAMIC DATA SYSTEM MODELING

Disadvantages of the traditional Dynamic Data System modoling method is first briefly described. The way to improve it is explained and the improved algorithm derived. Detailed block diagrams of the improved algorithm are given.First, a long AR model is used to fit the data untill the residuals are simply white noise. If the order of AR model is not high, the AR model is recognized.0therwhise,using the data and the residuals,the ARMA model parameters can be estimated by means of the linear Least Square Method. The coefficient matrices of Yule-Wolker equation and the model parameters can be recurrently calculated when the model order is increased or decreased. When preliminary estimation of suitable ARMA model is obtained, the nonlinaar Least Square method is used to optimize the para-meters.Two simulated examples are given and the results show that the improved modeling method can save ca.66%in time of computation attaining with the same precision.lt can be taken that the improved algorithm is a successful one.