铁道运输与经济2024，Vol.46Issue(2)：167-175,9.DOI:10.16668/j.cnki.issn.1003-1421.2024.02.21
多头图注意机制的深度学习地铁客流预测方法
Deep Learning Subway Passenger Flow Forecasting Method Based on Multi-head Graph Attention Mechanism
摘要
Abstract
Mining the spatial-temporal correlation of passenger flow among subway stations is a key to realizing high-precision subway passenger flow prediction.However,the spatial correlation of passenger flow between stations in the subway network is complex and difficult to quantify,which leads to excessive dependence on time correlation in passenger flow prediction.To overcome this defect,this paper proposed a deep-learning subway passenger flow forecasting method based on a multi-head graph attention mechanism.By constructing a multi-head graph attention mechanism oriented to the subway network,the paper studied the spatial relevance of passenger flow among multiple associated stations and obtained the differentiated weight value to quantify the spatial relevance of passenger flow between the target station and the associated station group.At the same time,the multi-head map attention mechanism was integrated into the long-term and short-term memory learning model(LSTM),and the passenger flow forecasting was realized by taking the quantified spatial correlation data as the input and combining the temporal correlation of the subway passenger flow.The experimental results show that the proposed method is feasible and effective,which can improve the prediction accuracy and the prediction results are better than those of the classical prediction methods.关键词
地铁/客流预测/图注意机制/长短时记忆神经网络/客流时空关联性Key words
Rail Transit/Passenger Flow Forecast/Graph Attention Network/LSTM/Spatial and Temporal Correlation of Passenger Flow分类
交通工程引用本文复制引用
张阳,陈燕玲..多头图注意机制的深度学习地铁客流预测方法[J].铁道运输与经济,2024,46(2):167-175,9.基金项目
国家自然科学基金项目(61976055) （61976055）
福建省自然科学基金项目(2023J01946) （2023J01946）
