一种用于季节性产品需求预测的多元化堆叠回归模型  

A Diversified Stacked Regression Model for Seasonal Product Demand Forecasting

作　　者：刘斌[1] 丁昊[1] LIU Bin;DING Hao(School of Management,University of Shanghai for Science&Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学管理学院,上海200093

出　　处：《物流技术》2024年第6期15-30,共16页Logistics Technology

基　　金：国家自然科学基金资助项目(71971134)。

摘　　要：产品需求预测是智慧供应链的核心环节。针对具有季节性的快消品的需求特点,设计了一种结合Blending线性与多机器学习模型融合的多元化堆叠回归模型RXOEL-X。首先,介绍了RXOEL-X模型的构建及运行步骤,然后基于一组公开数据将此模型与五种传统单一化模型进行比较,证明其在预测精度方面比其他模型更优。同时基于某饮料公司的实际销售数据,对模型性能进行进一步测试,证明RXOEL-X模型在预测精度、数据拟合能力、时间效率等方面整体表现最佳。RXOEL-X模型为季节性产品乃至更广泛的企业供应链管理中的需求预测问题提供了一种前沿的解决策略,有利于帮助企业在节省成本、减少库存积压的同时,提高对市场变化的响应速度和供应链的整体灵活性。With continuous changes in the global economic and industrial structure,effective supply chain management,especially accurate demand forecasting,has become the key for enterprises to cope with chal-lenges,avoid resource waste,reduce costs and improve operational efficiency.Therefore,it is of great signifi-cance to develop a forecasting model that is accurate,flexible and adaptable to market changes.In this paper,we proposed a diversified stacked regression model RXOEL-X,which combines the advan-tages of multiple algorithms including Blending Linear Regression,Random Forest(RF),Extreme Gradient Boost(XGBoost),Ordinary Least Squares(OLS),ElasticNet and Long Short-Term Memory Network(LSTM),and uses XGBoost as a secondary optimization model,which not only utilizes the powerful data analysis capabilities of machine learning,but also taps the robustness of traditional statistical methods and the nonlinear modeling capabilities of deep learning.The model fusion technology employed significantly im-proved the forecasting performance of the model,especially enabling it to effectively capture the seasonality and long-term dependence in the time series data,which is suitable for the demand forecasting of supply chains with obvious seasonality and trend characteristics.After introducing the construction and operation steps of RXOEL-X,the model is compared with five traditional simple models based on a set of public data,proving that the RXOEL-X model is better than the other models in terms of forecasting accuracy.At the same time,based on the actual sales data of a beverage company,the performance of the model was further tested and compared with 10 combination models,proving that the RXOEL-X model excelled in terms of prediction accuracy and data fitting ability.Through a sensitivity analysis,the forecasting accuracy of the RXOEL-X model was found to be virtually insusceptible to external influence,showing extremely high ro-bustness.In a temporal analysis,the model also performed best.The RXOEL-X model provides a frontier

关 键 词：季节性产品 需求预测 多元化堆叠回归模型 机器学习 智慧供应链 

分 类 号：F713.52[经济管理—市场营销]