| 摘 要: 准确快速的交通预测对于城市交通调控、路线规划和交通信号实时控制至关重要。目前提出的许多时空方法忽略了道路网络节点之间随着时间推移而产生变化的空间关联性。为解决这些问题,提出了一种基于改进交互式动态图卷积网络的交通流预测模型(I-IDGPSA)。该模型将交通数据划分为周期项和趋势项,通过交互式学习策略同步捕捉划分交通数据的时空相关性,并在后续步骤中加入概率稀疏自注意力机制,使得I-IDGPSA 相较于其他预测准确率相近模型的训练效率提升了9.07%,预测效率提升了52.13%。在 PeMS数据集上进行的大量实验结果表明,I-IDGPSA的性能优于最先进的基线方法。 |
| 关键词: 交互式学习 动态图卷积 时空相关性 概率稀疏自注意力机制 交通流预测 |
|
中图分类号: TP399
文献标识码: A
|
| 基金项目: “尖兵领雁+X”研发攻关计划(2024C01124) |
|
| ATraffic Flow Prediction Model Based on Improved Interactive Dynamic Graph Convolutional Network |
|
HU Xudong1, WU Baowen1,2, PENG Laihu1,2,QI Yubao1,2
|
(1.College of Mechanical Engineering, Zhejiang Sc-i Tech University, Hangzhou 310018, China;
2.Longgang Research Institute, Zhejiang Sc-i Tech University, Wenzhou 325000, China)
xdhu@zist.edu.cn; 202230503268@mails.zstu.edu.cn; laihup@zstu.edu.cn; 202020601042@mails.zstu.edu.cn
|
| Abstract: Accurate and rapid traffic prediction is crucial for urban traffic regulation, route planning, and rea-l time traffic signal control. Many existing spatiotemporal methods overlook the evolving spatial correlations among road network nodes over time. To address these issues, this study proposes a traffic flow prediction model based on an Improved Probabilistic Sparse self Attentive Interactive Dynamic Graph Convolutional Network (-I IDGPSA). The model divides traffic data into periodic and trend components, employs an interactive learning strategy to simultaneously capture the spatiotemporal correlations of the decomposed data, and incorporates a probabilistic sparse self attention mechanism in subsequent steps. As a result, I-IDGPSA improves training efficiency by 9. 07 % and prediction efficiency by 52.13% compared to other models with similar prediction accuracy. Extensive experimental results on the PeMS dataset demonstrate that -I IDGPSA outperforms state-o-f the-art baseline methods. |
| Keywords: ATraffic Flow Prediction Model Based on Improved Interactive Dynamic Graph Convolutional Network |
