基于改进Shapley值组合模型的航空货运量预测  

Air Cargo Volume Forecast Based on Improved Shapley Value Combination Model

作　　者：刘佳鑫 徐杭 王倩倩 白鸿宇 LIU Jiaxin;XU Hang;WANG Qianqian;BAI Hongyu(Zhejiang Provincial Department of Transport,Hangzhou 310009,China;Zhejiang Scientific Research Institute of Transport Transportation Development Research Center,Hangzhou 310023,China)

机构地区：[1]浙江省交通运输厅,浙江杭州310009 [2]浙江省交通运输科学研究院,浙江杭州310023

出　　处：《综合运输》2024年第8期132-137,共6页China Transportation Review

基　　金：浙江省交通运输厅科技计划项目(2023001);浙江未来交通科创中心发展研究专项(002202)。

摘　　要：航空货运是现代交通物流体系的重要组成部分,航空货运量预测对于发展规划和经营决策具有重要意义。本文在把握航空货运量影响因素及数据波动特征的基础上,选择多元回归模型、HoltWinters模型和RBF径向基神经网络模型,使用非线性优化模型改进Shapley值法,合理分配各预测模型的权重,基于此构建组合预测模型。提出的改进组合模型与选择的单项模型进行比较分析,结果表明,改进的组合模型的平均绝对百分比误差相比于单项预测模型误差分别降低了68.2%、20.4%、33.5%,相比于未改进的Shapley值法误差降低了1.4%,证明了改进组合模型的合理性及有效性,并利用该模型对中远期的航空货运量进行预测。Air cargo is an important part of modern transportation logistics system.Air cargo volume forecast is of great significance to development planning and management decision.On the basis of grasping the influencing factors of air cargo volume and the characteristics of data fluctuation,this paper selects the multivariate regression model,Holt-Winters model and RBF neural network model,and improves Shapley value method by the nonlinear optimization model to reasonably allocate the weight of each forecasting model.Based on this,the combined forecasting model is constructed.The improved combination model proposed in this paper is compared and analyzed with the selected single model.The results show that the average absolute percentage error of the improved combined model is reduced by 68.2%,20.4%and 33.5%compared with that of the single forecasting model,and reduced by 1.4%compared with the unimproved Shapley value method.The rationality and validity of the improved combination model are proved,and the model is used to forecast the air cargo volume in the medium and long term.

关 键 词：航空货运量 改进Shapley值 组合预测 多元回归模型 Holt-Winters模型 径向基神经网络 

分 类 号：U8[交通运输工程] F560.84[经济管理—产业经济]