期刊文献+

基于多图时空注意力的轨道交通客流预测模型 被引量:3

A Prediction Method of Rail Transit Passenger Flow Based on Multi-graph Spatial and Temporal Attention
下载PDF
导出
摘要 针对轨道交通客流由于复杂的时空相关性和显著的波动性而难以预测的问题,提出一种基于注意力机制和多图视角图神经网络的轨道交通客流量预测方法MGCNSTA。基于站点连接的邻接图和乘客出行的出发地-目的地(origin-destination, OD)图,采用长期和短期两种序列模式,通过图卷积神经网络和卷积神经网络分别捕获空间和时间序列特征,并运用注意力机制加强卷积模块的时空相关性特征。对杭州地铁短期客流量进行了预测,实验结果验证了模型的有效性。 Aiming at the problem that the rail transit passenger flow was difficult to predict due to the complex spatio-temporal correlation and significant volatility,a prediction method of rail transit passenger flow based on the multi-graph convolutional neural network for spatial and temporal attention was proposed to comprehensively predict the overall passenger flow of each subway station in a city.MGSTART was a multi-graph convolution method based on the adjacency graph of station connections and the origin-destination graph of passenger travel records.This method adopted two sequence modes(long-term sequence and recent sequence)to capture spatial features through the graph convolution neural network and time-series features through the convolution neural network.Meanwhile,it used an attention mechanism to strengthen the spatio-temporal correlation features of the convolution module.The model′s validity was verified by forecasting the short-term passenger flow of Hangzhou metro.Experiments showed that the MGSTART model outperforms the baseline model.
作者 陈俊彦 黄雪锋 韦俊宇 卢贤涛 卢小烨 CHEN Junyan;HUANG Xuefeng;WEI Junyu;LU Xiantao;LU Xiaoye(School of Computer Science and Information Security,Guilin University of Electronic Technology,Guilin 541004,China;Guangxi Cloud Security and Cloud Service Engineering Technology Research Center,Guilin 541004,China)
出处 《郑州大学学报(理学版)》 CAS 北大核心 2023年第4期39-45,共7页 Journal of Zhengzhou University:Natural Science Edition
基金 广西自然科学基金重点项目(2020GXNSFDA238001) 广西自然科学基金项目(2018GXNSFAA281318) 广西高校中青年教师科研基础能力提升项目(2020KY05033)。
关键词 轨道交通 注意力机制 多图视角 图神经网络 rail transit attention mechanism multi-graph graph neural network
  • 引文网络
  • 相关文献

参考文献3

二级参考文献4

共引文献75

同被引文献19

引证文献3

二级引证文献1

;
使用帮助 返回顶部