基于深度学习的物流需求预测模型研究  

Research on Logistics Demand Forecast Mode Based on Deep Learning

作　　者：李国祥 马文斌[1,2] 夏国恩 LI Guo-xiang;MA Wen-bin;XIA Guo-en(Department of Academic Affairs,Guangxi University of Finance and Economics,Nanning 530003,China;Guangxi Key Lab of Multi-source Information Miningand Security,Guangxi Normal University,Guilin 541004,China;Guangxi Key Laboratory Cultivation Base of Cross-border E-commerce Intelligent Information Processing,Guangxi University of Finance and Economics,Nanning 530003,China)

机构地区：[1]广西财经学院教务处,广西南宁530003 [2]广西师范大学广西多源信息挖掘与安全重点实验室,广西桂林541004 [3]广西财经学院广西跨境电商智能信息处理重点实验室培育基地,广西南宁530003

出　　处：《系统科学学报》2021年第2期85-89,共5页Chinese Journal of Systems Science

基　　金：国家自然科学基金(71862003);广西重点研发计划(2018AB15003);广西跨境电商智能信息处理重点实验室培育基地(广西财经学院)专项资助项目;广西多源信息挖掘与安全重点实验室开放基金项目(MIMS17-02);广西高校中青年教师基础能力提升项目(2018KY0520);广西财经学院青年教师科研发展基金(2018QNA02)。

摘　　要：新一轮的科技革命和经济结构的不断优化对于区域物流需求的预测提出了更高的要求。本文以LSTM为研究对象,提出了基于深度学习的区域物流预测模型。从预测模型数据结构的角度,在时间序列和影响因子两个层面上,将深度学习预测模型与传统统计学方法,神经网络方法,基于优化寻参的支持向量回归机等方法进行综合对比评估。仿真结果表明,基于时间序列数据源的预测模型普遍优于影响因子数据源模型,而基于LSTM区域物流预测模型具有良好的预测性能,平均绝对百分误差4%左右,且稳定性好。With the revolution of technology and economic structure optimization, higher requirements are put forward for the prediction of regional logistics demand. In this paper, a regional logistics forecasting model based on LSTM deep neural network is proposed. There are two data structures in the prediction model: time series and impact factors.Based on the above two data structures, different methods are compared, such as raditional statistical methods, neural network methods and optimized support vector regression.The experimental results show that prediction models based on time series are generally superior to impact factor data models, and the LSTM deep neural network can control the prediction error within 4%withstability.

关 键 词：物流预测 时间序列 影响因子 LSTM 

分 类 号：N949[自然科学总论—系统科学]