产品订单分析与需求预测  

Product Order Analysis and Demand Forecasting

作　　者：郝青松 卢冬晖 谭奥成 吴昊天 HAO Qingsong;LU Donghui;TAN Aocheng;WU Haotian(School of Mechanical Engineering,Wuhan Polytechnic University,Wuhan,Hubei 430000,China;Public Course Department,Xiangyang Polytechnic,Xiangyang,Hubei 441000,China)

机构地区：[1]武汉轻工大学机械工程学院,湖北武汉430000 [2]襄阳职业技术学院公共课部,湖北襄阳441000

出　　处：《数学建模及其应用》2023年第4期84-94,共11页Mathematical Modeling and Its Applications

摘　　要：需求预测是企业供应链管理的基础,高效、准确的需求预测有助于采购计划和生产计划制定,减少业务波动影响.某企业相关数据具有时序性、非线性、时间跨度大等特性.对数据预处理后,进行特征工程处理,使用MMIFS算法来量化各个特征与订单需求量间相关性.使用CRU模型、DeepAR模型和Prophet模型来针对不同时间粒度需求量建模,以时间滚动交叉检验RMSE作为模型评估标准.不规则时间序列数据,不同时间粒度趋势、波动、突变不同,适用于不同模型.交叉检验评估验证了模型有效性,表明本文方法能够有效运用于其他不规则时序需求预测问题.Demand forecasting is the foundation of enterprise supply chain management.Efficient and accurate demand forecasting helps in procurement and production planning,reducing the impact of business fluctuations.The relevant data of a certain company has characteristics such as time dependence,non-linearity,and a large time span.After data preprocessing,feature engineering is performed,and the MMIFS algorithm is used to quantify the correlation between each feature and the order demand.The CRU model,DeepAR model,and Prophet model are used to model the demand at different time granularities,with time-rolling cross-validation Mean Squared Error(RMSE)as the model evaluation criteria.Irregular time series data exhibit different trends,fluctuations,and mutations at different time granularities,thus requiring the application of different models.Cross-validation evaluation confirms the effectiveness of the models,indicating that the method proposed in this paper can be effectively applied to other irregular time series demand forecasting problems.

关 键 词：不规则时间序列 BGCP插补 CRU模型 DeepAR模型 Prophet模型 

分 类 号：O29[理学—应用数学]