摘要
本文研究了非平稳时间序列的状态空间建模与预测方法.在建模的过程中,采用了分解模型的方法,从原始时间序列直接建立起状态空间模型,对趋势性建立起随机走动模型,对周期性因素建立了动态谐波模型,从而摆脱了Box—Jenkins建立ARIMA模型的繁琐程序.
The thesis deals with state space modeling and forecasting of time series. Decompostion model methods is proposed to construct the model, ba: d on this method, the original time series can be modeled directly into .afate space formula. The trend component is modeled as random - walk model, the periodical component is modeled as dynamic harmonic regression model. Finally, the complete state space model is formed,this method removes the complexity in AR1MA model proposed by box and Jenkins.
出处
《系统工程》
CSCD
1998年第3期54-59,69,共7页
Systems Engineering
关键词
时间序列
状态空间模型
EM算法
参数估计
Time series, State space model, EM algorithm, Kalman Filtering, Theta Algorithm, Optimal fixed interval smoothing