基于周期项方法选择的季节性时序预测  

Seasonal Time Series Forecasting Based on Seasonality Method Selection

作　　者：宋仙磊[1,2] 刘业政[1,2] 陈思凤[1,2] 

机构地区：[1]合肥工业大学管理学院,合肥230009 [2]合肥工业大学过程优化与智能决策教育部重点实验室,合肥230009

出　　处：《计算机工程》2011年第21期131-132,135,共3页Computer Engineering

基　　金：国家自然科学基金资助项目(71071047);高等学校博士点基金资助项目(20090111110016)

摘　　要：根据每个单步预测序列各自具有的特征,通过周期项重构把多步预测转化为单步预测,提出一种预测方法选择策略。为每个单步预测序列选择一个最合适的预测方法,利用选择的方法建模预测周期项,结合灰色预测模型对趋势项的预测值,建立季节性时间序列整体预测模型。实验结果表明,该模型能克服周期项多步预测的缺点,具有较高的预测精度。The seasonality of seasonal time series is reconstructed to transform the multi-step ahead forecasting into a single-step forecasting.According to the characteristics of every single-step forecasting time series,a forecasting selection approach is presented.As for every single-step forecasting,most proper forecasting method comes up,then the method selected is used to build a model to predict seasonality.Combining the forecasted trend with the predicted values obtained by a grey forecasting model,the integral seasonal time series forecasting model is established.The comparison of forecasting results show that this model outperforms the multi-step ahead forecasting with better forecasting performance.

关 键 词：周期项重构 方法选择 周期项预测 季节性时间序列 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]