基于多重“分解—集成”策略的物流货运量预测  

Logistics Freight Volume Forecasting Based on Multilevel Decompose-ensemble Method

作　　者：周程[1] 李松[2] 

机构地区：[1]湖北经济学院物流与工程管理学院,武汉430205 [2]武汉理工大学物流工程学院,武汉430063

出　　处：《交通运输系统工程与信息》2015年第1期150-158,共9页Journal of Transportation Systems Engineering and Information Technology

基　　金：国家社会科学基金项目(14BJY139);国家自然科学基金项目(51175394);湖北省教育厅人文社科(2012Q099);湖北物流发展研究中心资助项目(2014A03)

摘　　要：货运量预测是制定物流政策和决定物流基础设施布局的重要依据.针对受多因素影响的货运量预测具备较强非线性和模糊性特征,提出一种基于趋势分解和小波变换的多重'分解—集成'预测方法.利用趋势分解将货运量分解为趋势项和非趋势项,通过小波分解将非趋势项进一步分解成低频项和高频项,分别建立预测模型,选用相加集成得到货运量预测值.实证表明,'分解—集成'的预测策略将非平稳货运量分解为相对平稳的子序列组合,降低了问题复杂度,有效提高了预测性能,与传统的趋势分解预测模型和小波分解预测模型相比,多重'分解—集成'预测模型精度更高.Logistics freight volume forecasting is essential for forming logistics policy and determining the logistics infrastructure layout, which reflects strong-nonlinearity and ambiguity due to various affecting factors. A new forecasting approach based on multilevel decompose-ensemble is proposed for logistics freight volume. Original freight volume is firstly decomposed into trend component and non-trend component in accordance with trend decomposition. Then, non-trend component is further decomposed into a low frequency subseries and a several high frequency subseries by using of wavelet decomposition. With respect to their different features, trend component, low frequency non-trend component and high frequency non-trend component are respective forecasted. The prediction result of freight volume is the superimposition of these subseries predictions. Non-stationary time series is resolved into relatively stationary subsequences in accordance with trend decomposition and wavelet decomposition. The empirical test proves that the proposed forecasting method based on multilevel decompose-ensemble method is higher accuracy, which is compared with traditional decompose-ensemble forecasting method based on trend decomposition or wavelet decomposition.

关 键 词：物流工程 小波变换 趋势分解 分解—集成 物流货运量 

分 类 号：F542[经济管理—产业经济]