基于单变量分解集成的牛奶消费需求预测研究  

UNIVARIATE DECOMPOSE-ENSEMBLE METHOD BASED MILK DEMAND FORECASTING

作　　者：王帅[1] 汤铃[2] 余乐安[2] 

机构地区：[1]中国科学院数学与系统科学研究院国家数学与交叉科学中心,北京100190 [2]北京化工大学经济管理学院,北京100029

出　　处：《系统科学与数学》2013年第1期11-19,共9页Journal of Systems Science and Mathematical Sciences

基　　金：国家科技支撑计划重点项目"农产品数量安全智能分析与预警关键技术支撑系统及示范";国家杰出青年科学基金(71025005);国家自然科学基金重大研究计划培育项目(90924024)及国家自然科学基金重点项目(91224001)资助课题

摘　　要：牛奶消费需求预测对牛奶价格的稳定以及奶业生产的计划安排、销售决策具有重要意义.选取牛奶的全国年度总消费量作为研究对象,提出基于集合经验模态分解(Ensemble Empirical Mode Decomposition,EEMD)/小波(Wavelet)分解和最小二乘支持向量回归(Least Squares Support Vector Regression,LSSVR)的单变量分解集成方法,以对牛奶消费需求量进行预测研究.实证检验表明,所提出的单变量分解集成预测方法相比单一预测模型能更为有效地预测牛奶消费需求.外推预测结果显示:2010-2012年我国牛奶消费量将呈现出上升的趋势.牛奶预测精度的有效提高将有助于有关决策部门提前做好调控工作,从而保证奶业市场的健康发展.Prediction for future market demand of milk is important for stabilizing milk price,developing marketing strategies and production planning decisions.This paper proposes a novel univariate decompose-ensemble methodology which uses the ensemble empirical mode decomposition（EEMD）,wavelet decomposition,and least squares support vector regression （LSSVR） to predict milk consumption in China.At the same time,the single LSSVR method is applied for comparison purpose.In the decompose-ensemble methods,EEMD and wavelet decomposition methods are first used to decompose the original data and then LSSVR approach is used to predict the separated components.Finally,the prediction results of different components are combined to formulate the ensemble result.It can be seen from the forecasting results that the milk demand from 2010 to 2012 will increase.Based on the result,the related departments should take actions to ensure the healthy development of dairy market in China.

关 键 词：牛奶消费需求量 WAVELET EEMD LSSVR 预测 

分 类 号：F326.3[经济管理—产业经济] F224