基于迁移学习的新产品销量预测模型  

Sales Forecasting Model of New Products Based on Transfer Learning

作　　者：武玉英[1,2] 孙平 何喜军 蒋国瑞[1,2] WU Yu-ying;SUN Ping;HE Xi-jun;JIANG Guo-rui(Collede of Economics and Management,Beijing University of Technology,Beijing 100124,China;Research Base of Beijing Modem Manufacturing Development,Beijing 100124,China)

机构地区：[1]北京工业大学经济与管理学院,北京100124 [2]北京现代制造业发展研究基地,北京100124

出　　处：《系统工程》2018年第6期124-132,共9页Systems Engineering

基　　金：国家自然科学基金资助项目(71371018);北京市社会科学基金资助项目(15JGB124);北京市自然科学基金资助项目(9172002);北京工业大学研究生科技基金资助项目(ykj-2017-00437)

摘　　要：本文提出一种基于迁移学习的销量预测模型,以应对小样本数据下新产品销量预测问题。首先,对候选产品集与新产品进行多维特征相似比较,确定源域产品;然后,基于样本迁移的思想,利用类比合成法,在源域时间序列中匹配出与目标域时间序列高相关的多个模式,并采用最小二乘法和遗传算法筛选最佳模式,确定模式长度,以进行新产品的联合销量预测。实验结果表明:与其他经典预测方法相比,该方法能提高新产品销量预测准确度,并验证了模型的合理性和科学性,为新产品销量预测提供了有效思路。A sales forecasting model based on transfer learning is proposed to deal with the problem of new product sales forecasting under small sample data.First,the candidate product set is compared with the new product in terms of multi- dimensional characteristics to determine the source domain product.Then,based on the idea of sample transfer,we use the analog complexing method to match the multiple patterns which are highly related to the time series of the target domain in the source domain time series,and select the best pattern by using the least square method and genetic algorithm to determine the pattern length so as to forecast the joint sales of the new product.The results show that compared with those of other classical algorithms, this method can improve the accuracy of new product sales forecasting,which validates the rationality and scientificity of the model,and provides an effective way to forecast the sales of new products.The results show that compared with those of other classical algorithms,this method can improve the accuracy of new product sales forecasting,which validates the rationality and scientificity of the model,and provides an effective way to forecast the sales of new products.

关 键 词：销量预测 新产品 迁移学习 类比合成法 遗传算法 

分 类 号：F224[经济管理—国民经济]