Abstract: With the development of industrial Internet economy,the uncertainty of demand is increasing.To improve the accuracy of demand forecasting,a hierarchical Bayesian Demand Forecasting model with Side Information (DFSI) was proposed,which could realize the dynamic description of customer demands by its latent tier.The latent tier mainly consisted of two sets of parameters,one set was used to describe the intrinsic demand continuity over time,and the other set was used to interplay with the side information.In addition to the model,the optimization target of DFSI was derived,and a corresponding solution algorithm was designed as well.For evaluation,the comprehensive experiments with three real datasets were conducted,and all the results showed that the proposed method outperformed the other state-of-the-art algorithms.

Key words: supply chains, demand forecasting, side information, nonparametric model, Bayesian inference

摘要： 随着工业互联网经济的发展,需求的不确定性日益增大,为提高需求预测的准确性,提出一个考虑边信息的多层贝叶斯需求预测模型(DFSI)。DFSI模型通过构造隐层的网络结构以实现对客户需求更加精确的刻画,该隐层结构主要包含两组参数：一组用于描述客户需求在时间上固有的连续性特征,另一组则用于融合相关的边信息特征。进一步,以贝叶斯推断为理论基础,以最大化后验概率为目标,推导出了DFSI的优化目标,并基于梯度下降方法设计了相应的求解算法。使用京东商城及某制造企业的真实销售数据对提出的模型进行了检验。结果显示,与常用的需求预测模型相比,DFSI能获得更好的预测结果。

关键词: 供应链, 需求预测, 边信息, 非参数模型, 贝叶斯推断