基于CBP的卷烟销售二重趋势时间序列预测模型研究与应用.pdfVIP

第24卷第6期 控 制 理 论 与 应 用 、，0l_24 NO．6 2007年l2月 Control Theory Applications Dec．2007 文章编号：1000—8152(2007)06—1015—06 基于CBP的卷烟销售二重趋势时问序列预测模型研究与应用 谢星峰 ，谢东风 ，邹 平。 (1．昆明理工大学管理经济学院，云南昆明650093；2．云南烟草公司，云南昆明65001 1 3．云南师范大学，云南 昆明650092) 摘要：二重趋势时间序列的预测结果与其整体趋势变动性和季节波动性的特征有较大的关系，在采用神经网络 并针对其特征的基础上还没有较好的预测方法．本文以圆形反向传播网络(CBP)为基础，提出二重趋势时间序列预 测组合模型(CCBP)，即根据二重趋势时间序列特征对其趋势分量与周期分量采用圆形反向传播网络分开进行预测， 再对二者积乘．以云南省某片区若干卷烟销售网点的采样数据为研究实例，证实了CCBP模型在二重趋势时间序列 预测方面性能有所提高． 关键词：圆形反向传播网络；二重趋势时间序列：预测 中图分类号：TP273 文献标识码：A Study and application of double trend time series prediction models of cigarette sell based on CBP XIE Xing—feng ，XIE Dong—feng ，ZOU Ping (1．School of Management and Economy,Kunming University of Science and Technology， Kunming Yunnan 650093，China； 2．Yunnan Province Tobacco Corporation，Kunming Yunnan 65001 1，China； 3．Yunnan Normal University,Kunming Yunnan 650092，China) Abstract：The prediction performance of double trend time series has dependence relation to its character：the trend change nature and season fluctuation characteristic．However,there iS no effective way to construct the forecast method with the Neural Network and this double trend character．Based on circular back propagation network(CBP)，a combined forecast model-CCBP is proposed which separates the weight of trend and the weight of fluctuation from predicted series data，respectively predicts the two weights，then multiplies two weight．After comparatively studying the data cases of some Yunnan tobacco company,it is proved that the CCBP model in the