交通运输系统工程与信息

交通运输系统工程与信息 ›› 2025, Vol. 25 ›› Issue (2): 169-179.DOI: 10.16097/j.cnki.1009-6744.2025.02.016

• 系统工程理论与方法 • 上一篇    下一篇

基于扩散模型的节假日高速公路交通流预测方法

林培群1,2，陈泽沐1,2，周楚昊*1,2   

  1. 1. 华南理工大学，土木与交通学院，广州510641；2.中新国际联合研究院，广州510555
  • 收稿日期:2024-12-07 修回日期:2025-02-08 接受日期:2025-02-14 出版日期:2025-04-25 发布日期:2025-04-20
  • 作者简介:林培群(1980—)，男，广东饶平人，教授，博士。
  • 基金资助:
    国家自然科学基金(52072130)；广东省科技计划项目(2023A1111120018)。

Expressway Traffic Flow Prediction Method for Holidays Based on Diffusion Model

LIN Peiqun1,2,CHEN Zemu1,2,ZHOU Chuhao*1,2   

  1. 1. School of Civil Engineering and Transportation, South China University of Technology, Guangzhou 510641, China; 2. China-Singapore International Joint Research Institute, Guangzhou 510555, China
  • Received:2024-12-07 Revised:2025-02-08 Accepted:2025-02-14 Online:2025-04-25 Published:2025-04-20
  • Supported by:
    National Natural Science Foundation of China (52072130)；Guangdong Provincial Applied Science and Technology Research and Development Program (2023A1111120018)。

PDF

摘要： 交通管理部门需要依赖精准的交通需求预测来制定科学的交通管控措施。然而，节假日期间交通流的高度不确定性和突发性使得在假期开始前进行精确预测变得困难。本文基于扩散概率模型理论，构建一种适用于节假日交通流预测的扩散框架，并进一步开发了一个融合流量时空特征、节假日特性和天气因素的多特征提取条件扩散模型(ConditionalDiffusionModelwith Multi-feature Extraction, CDMME)，以实现节假日高速公路路段的长时交通流预测。本文选取广东省内28段繁忙高速公路路段在元旦、端午节和中秋节假期的15min及1h流量数据进行模型训练和测试。实验结果表明，与随机森林模型相比，CDMME在15min和1h总流量预测任务中的加权平均绝对误差百分比(WeightedMeanAbsolutePercentage Error, WMAPE)分别下降了12.98%和34.88%，平均绝对误差(MeanAbsolute Error, MAE)分别增大了1.47%和下降了23.54%；与长短时记忆网络模型相比，WMAPE分别降低了16.10%和32.39%，MAE分别降低了9.42%和27.55%。与15 min总流量预测、1h客车流量预测和1h货车流量预测相比，1h总流量预测的WMAPE分别下降了29.57%、12.23%和30.42%，表明在数据量级更大的任务中，CDMME的性能更为优越。可视化结果显示，CDMME在捕捉流量峰值方面表现更出色。此外，提前1天进行预测时，CDMME的平均预测准确率最高，1h总流量预测的精度可达到87.27%。

关键词: 智能交通, 长时交通流预测, 条件扩散模型, 高速公路, 节假日

Abstract: Traffic management authorities require accurate traffic demand forecasts to implement effective traffic control strategies. However, the high uncertainty and suddenness of holiday traffic flows present significant challenges in generating precise pre holiday predictions. This research introduces a diffusion framework for predicting holiday traffic flow, grounded in diffusion probabilistic model theory, and further develops a Conditional Diffusion Model with Multi-feature Extraction (CDMME). The proposed CDMME integrates spatio-temporal characteristics of traffic flow, holiday attributes and meteorological factors to predict long-term traffic flow for holidays. Experiments are conducted using 15-minute and one-hour traffic flow data from 28 busy expressway segments in Guangdong Province, focusing on holidays such as New Year's Day, the Dragon Boat Festival and the Mid-Autumn Festival for model training and validation. The experimental results indicate that, for 15-minute and hourly total flow predictions, compared to the random forest (RF) model, the CDMME reduces the Weighted Mean Absolute Percentage Error (WMAPE) by 12.98% and 34.88%, respectively, while the Mean Absolute Error (MAE) increases by 1.47% for 15-minute prediction and decreases by 23.54% for hourly prediction. In comparison to the long short-term memory (LSTM) model, the CDMME reduces the WMAPE by 16.10% and 32.39% , respectively, and the MAE by 9.42% and 27.55% respectively. Additionally, when comparing hourly total traffic prediction with 15-minute total traffic prediction, hourly passenger traffic prediction and hourly truck traffic prediction, the WMAPE decreased by 29.57%, 12.23% and 30.42%, respectively, indicating that it has superiority in tasks with larger magnitude. Visualization result demonstrates that the CDMME effectively captures traffic peaks. Furthermore, the CDMME achieves peak average accuracy with a one-day advance forecast, with the accuracy of hourly total traffic prediction reaching 87.27%.

Key words: intelligent transportation, long-term traffic flow prediction, conditional diffusion model, expressway, holiday

中图分类号: