铁道科学与工程学报2025，Vol.22Issue(10)：4436-4448,13.DOI:10.19713/j.cnki.43-1423/u.T20250124
基于自适应时空分层网络的城市轨道交通站点客流预测
Urban rail transit station passenger flow prediction based on adaptive spatio-temporal hierarchical network
摘要
Abstract
Accurate passenger flow prediction for urban rail transit is crucial for rail transit network planning,intelligent travel,and operation management.Current passenger flow prediction methods have made certain progress in the extraction of spatio-temporal features,but there are still limitations in the integration of inherent patterns such as spatial distribution,topological relevance,and temporal periodicity.To address the problem of balancing dynamic changes and long-term trends when precisely constructing spatio-temporal dependencies,a passenger flow prediction method based on the adaptive spatio-temporal hierarchical network(ASTHN)was proposed.The innovations of this method were mainly reflected in two aspects:First,a temporal aggregation module was designed to learn the periodicity and trend of passenger flow through an adaptive mechanism,enhancing the model's ability to understand and remember passenger flow patterns in different periods of rail transit,and compensating for the insufficient recognition ability of traditional temporal embedding for temporal features.Second,a fully symmetric spatio-temporal encoding and feature reconstruction module was constructed,combining deep learning model components such as dilated convolution,gating mechanisms,and adaptive graph convolution.The spatio-temporal encoding module extracted key features from the current passenger flow information to form a high-dimensional feature expansion structure,deeply mining the hidden states and inherent patterns in the passenger flow data.The feature reconstruction module mapped historical hidden states and reconstructed future hidden states to achieve comprehensive representation of future features.Experimental results based on Hangzhou Metro passenger flow data show that the prediction accuracy of ASTHN is higher than that of existing baseline models.Visualization analysis verifies its ability to capture complex passenger flow patterns,and ablation experiments prove the necessity of each module.The proposed method can provide reliable decision support for the intelligent management of rail transit,and assist in passenger flow regulation and capacity optimization.关键词
客流预测/城市轨道交通/自适应时空分层网络/时空特征重构/自适应图卷积Key words
passenger flow prediction/urban rail transit/adaptive spatio-temporal hierarchical network/spatio-temporal feature reconstruction/adaptive graph convolution分类
交通运输引用本文复制引用
曾璐,江子璇,彭东良,颜树成..基于自适应时空分层网络的城市轨道交通站点客流预测[J].铁道科学与工程学报,2025,22(10):4436-4448,13.基金项目
"十四五"国家重点研发计划项目(2023YFB4302100) （2023YFB4302100）
江西省重大科技研发专项项目(20232ACE01011) （20232ACE01011）
磁浮轨道交通装备江西省重点实验室资助项目(2020SSY050) （2020SSY050）
