%0 Journal Article
%T 基于自适应时空图交互的城市轨道交通短时客流预测
Short-Term Passenger Flow Prediction for Urban Rail Transit Based on Adaptive Spatiotemporal Graph Interaction
%A 颜树成
%A 王昌友
%A 江子璇
%J Journal of Sensor Technology and Application
%P 480-489
%@ 2331-0243
%D 2025
%I Hans Publishing
%R 10.12677/jsta.2025.133048
%X 城市轨道交通客流预测对于地铁运营调度、客流管理及个人出行规划具有重要意义。然而，由于不同站点的流量模式各异，以及站点间的空间依赖关系具有时变性，该问题仍然面临诸多挑战。为此，本文提出了一种自适应时空图交互模型(ASTGN)，用于精准预测城市轨道交通流量。首先，引入异构节点嵌入(HNE)模块，以捕获不同站点的个性化流量模式。其次，采用动态图依赖建模(DGDM)模块，从数据中自适应学习站点间的动态空间关系，无需预定义邻接结构。然后，利用多步时序注意力(MSTA)模块，基于自注意力机制建模长期时间依赖关系，以提升长时序预测能力。最后，利用动态融合模块(DFM)整合时空信息，使模型更好地适应复杂的轨道交通流量变化。在杭州地铁数据集上进行了实验，并与10种基线方法进行对比，结果表明ASTGN在多个时间粒度上的预测精度均显著优于现有方法，能够有效提升城市轨道交通客流预测的可靠性与稳定性。
Urban rail transit passenger flow prediction plays a crucial role in metro operation scheduling, passenger flow management, and individual travel planning. However, due to the diverse traffic patterns across different stations and the time-varying spatial dependencies between stations, this problem remains highly challenging. To address these challenges, this paper proposes an Adaptive Spatiotemporal Graph Interaction Model (ASTGN) for accurate urban rail transit flow prediction. First, the Heterogeneous Node Embedding (HNE) module is introduced to capture the personalized traffic patterns of different stations. Second, the Dynamic Graph Dependency Modeling (DGDM) module is employed to adaptively learn the dynamic spatial relationships between stations from data without requiring predefined adjacency structures. Third, the Multi-Step Temporal Attention (MSTA) module, based on a self-attention mechanism, is used to model long-term temporal dependencies and enhance long-range prediction capabilities. Finally, the Dynamic Fusion Module (DFM) integrates spatial and temporal information to enable the model to better adapt to complex rail transit flow variations. Extensive experiments on the Hangzhou metro dataset, compared against ten baseline methods, demonstrate that ASTGN consistently achieves superior predictive accuracy across multiple time granularities, significantly enhancing the reliability and stability of urban rail transit passenger flow forecasting.
%K 客流预测，
%K 时空数据，
%K 图神经网络，
%K 深度学习
Passenger Flow Prediction
%K Spatiotemporal Data
%K Graph Neural Network
%K Deep Learning
%U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=115411