交通运输系统工程与信息

交通运输系统工程与信息 ›› 2026, Vol. 26 ›› Issue (2): 300-308.DOI: 10.16097/j.cnki.1009-6744.2026.02.028

• 工程应用与案例分析 • 上一篇    下一篇

融合贝叶斯优化与深度学习的机场公交专线短时客流预测

刘涛*1a,1b,2 ，李林1a,2   

  1. 1. 西南交通大学，a.交通运输与物流学院，b.综合交通大数据应用技术国家工程实验室，成都611756； 2. 宜宾西南交通大学研究院，四川宜宾644000
  • 收稿日期:2025-11-14 修回日期:2026-01-15 接受日期:2026-01-26 出版日期:2026-04-25 发布日期:2026-04-21
  • 作者简介:刘涛（1989—），男，四川人，教授，博士。
  • 基金资助:
    国家自然科学基金 (72271206)；宜宾市科技计划项目(2024JC002)。

Short-Term Passenger Flow Prediction for Airport Express Buses by Integrating Bayesian Optimization and Deep Learning Models

LIU Tao*1a,1b,2, LI Lin1a,2   

  1. 1a. School of Transportation and Logistics, 1b. National Engineering Laboratory of Integrated Transportation Big Data Application Technology, Southwest Jiaotong University, Chengdu 611756, China; 2. Yibin Research Institute of Southwest Jiaotong University, Yibin 644000, Sichuan, China
  • Received:2025-11-14 Revised:2026-01-15 Accepted:2026-01-26 Online:2026-04-25 Published:2026-04-21
  • Supported by:
    National Natural Science Foundation of China (72271206)；Yibin Municipal Science and Technology Project (2024JC002)。

PDF

摘要： 针对机场公交专线客流具有强外部干扰性和高随机性特征，本文采用一种基于贝叶斯优化的混合深度学习模型进行客流短时预测。该模型通过卷积神经网络提取客流数据的局部特征，引入注意力机制实现关键时刻动态加权，从而增强模型预测复杂客流的能力，利用长短期记忆网络捕捉长期的时序依赖关系。为克服传统人工调参的局限性，采用贝叶斯优化算法自动搜索全局最优超参数组合，以增强模型的预测精确性。本文以成都天府国际机场公交专线4号线2024年1月—10月的客流数据为例，分析5、10、15、20 min这4种时间粒度对预测精度的影响。实验结果表明，在20min时间粒度下，模型的拟合优度达到0.794 2，均方根误差为3.836 9，平均绝对误差为3.132 3，加权平均绝对百分比误差为19.041 8%。与Transformer、随机森林、XGBoost、时空同步图卷积网络（STSGCN）、图多重注意力网络（GMAN）和时空注意力融合网络（STAFN）等基准模型相比，本文所提出的模型在各项指标上均有一定提升，验证了模型在机场公交专线短时客流预测中的有效性与优越性。

关键词: 智能交通, 短时客流预测, 深度学习, 机场公交专线

Abstract: Considering the significant external disturbances and high randomness of airport express bus passenger flow, this study proposes a hybrid deep learning model for forecasting the short-term passenger flow. This model employs a convolutional neural network to extract the local features from passenger flow data. Then it introduces an attention mechanism to dynamically weight critical time steps, which significantly enhances its capability to predict complex passenger flow patterns. And it utilizes a long short-term memory network to capture long-term temporal dependencies. To address the limitations of manual hyperparameter tuning, a Bayesian optimization algorithm is employed to automatically search for the global optimal hyperparameter combinations, which significantly improves the prediction accuracy of this model. This paper takes the passenger booking data from Line 4, which collected from January to October 2024, an airport express bus service operating at Chengdu Tianfu International Airport as an example. Four temporal granularities (5, 10, 15, 20 minutes) are analyzed to assess their effect on prediction accuracy. The experimental results show that at a 20-minute granularity, the model achieves the best overall performance, with an R2 of 0.794 2, root mean square error (RMSE) of 3.836 9, mean absolute error (MAE) of 3.132 3, and mean absolute percentage error (WMAPE) of 19.041 8%. Compared to benchmark models like Transformer, Random Forest, XGBoost, spatial-temporal synchronous graph convolutional network (STSGCN), graph multi-attention network (GMAN), and spatial temporal attention fusion network (STAFN), the proposed model demonstrates a better prediction performance. This validates its effectiveness and advantages in short-term passenger flow forecasting for airport express bus services.

Key words: intelligent transportation, short-term passenger flow forecasting, deep learning, airport express bus

中图分类号: