摘 要：在城市公交客流预测中，传统方法难以同时有效捕捉空间动态变化与时间长期依赖之间的耦合关系。为获得更准确的预测结果，本文提出一种基于时空图神经网络的GraphTST模型。该模型融合了全局上下文嵌入机制、自适应动态图卷积编码器，以及基于Patch的Transformer时序模块。在真实公交数据集上的实验结果显示，GraphTST模型在多步预测任务中的平均绝对误差和均方根误差分别比Informer模型降低16%和约19%，表明该模型在预测精度与鲁棒性方面具有显著优势。
关键词：客流预测；时空图神经网络；动态图卷积；Transformer
中图分类号：U469.1+4；TP391 文献标志码：A DOI：10.15917/j.cnki.1006-3331.2026.03.006
Urban Bus Passenger Flow Prediction Method Based on Spatio-temporal Graph Neural Network
YANG Wenxin, HUANG Zhengxiang, XIAO Meng, LUO Si, ZHONG Zhikang, HE Dan
Abstract: In urban bus passenger flow prediction,traditional methods struggle to simultaneously capture the coupling between dynamic spatial variations and long-term temporal dependencies.To achieve more accurate prediction results,this paper proposes a GraphTST model based on a spatio-temporal graph neural network.The model integrates a global context embedding mechanism,an adaptive dynamic graph convolutional encoder,and a Patch-based Transformer temporal module.Experimental results on real bus datasets show that the GraphTST model reduces the mean absolute error and root mean square error in multi-step prediction tasks by 16%and above 19%,respectively compared to the Informer model,demonstrating that the proposed model has significant advantages in prediction accuracy and robustness.
Key words: passenger flow prediction; spatio-temporal graph neural network; dynamic graph convolution; Transformer
