摘要
现实中的时序数据,往往取自于复杂系统,表现出长记忆效应与短时不规则波动同时并存。传统的时序数据的分析和预测方法一般对不同层次的影响不加以区分,而是为其建立一个统一的模型,这使得在对复杂系统建模时需要用大量的参数予以表征,影响预测效率与精度。为此采用新的方法,将序列数据本身进行多平滑因子分解,对分解后的序列进行多尺度的采样并分别建模、预测,最后将结果整合。该方法应用于股票的实验表明,即使对起伏波动很大的时间序列,也能够得到较好的预测结果。
Time series are often produced in complex systems which are controlled both by macroscopic level and microscopic level laws, with long memory effect and short-term irregularly fluctuations coexisting in the same series. Traditional analysis and forecasting methods didn＇t distinguish these multi-level influences and always made a single model for predication, which had to introduce a lot of parameters to describe the characteristics of complex systems and result in the loss of efficiency and accuracy. However,we decomposed time series into several ones with different smoothness, all the sub-time series were respectively modeled and predicated with multi-scale sampling. Then the forecasting results of sub-time series were composed to get the result of the original time series. The experiment results on the stock forecasting show that the method is efficient, even for the time series with large fluctuations.
出处
《计算机应用》
CSCD
北大核心
2006年第4期888-890,894,共4页
journal of Computer Applications
基金
国家自然科学基金资助项目(60402010)
航天基金资助项目(No.2003HTZJDX13)
关键词
复杂系统
时间序列预测
多尺度采样
序列分解
complex systems
time series forecasting
multi-scale sampling
time series decomposing
