| 摘 要: 交通流预测是提高路网利用效率,缓解城市交通拥堵问题的重要方法之一。为了进一步提高交通流预测精度,提出了一种基于时间信息增强Transformer的短时交通流预测模型。该模型利用多图卷积网络从不同角度建模空间特征,利用长短期记忆网络和Transformer同时建模短期时间特征和长期时间特征。此外,模型采用了一种轻量化的结构以提高模型实时响应速度。在PEMS(Performance Measurement System)数据集上与基线方法相比,该模型的预测精度提高了5%,时间复杂度降低了65%。实验结果表明,基于时间信息增强Transformer的短时交通流预测模型在有效提取交通流数据中时空特征的同时,显著降低了模型的复杂度。 |
| 关键词: 交通流预测;深度学习;长短期记忆网络;Transformer;图卷积网络 |
|
中图分类号: TP399
文献标识码: A
|
| 基金项目: 国家自然科学基金项目(61603154);浙江省自然科学基金资助项目(LTGS23F030002);嘉兴市应用性基础研究项目(2023AY11034);工业控制技术国家重点实验室开放课题(ICT2022B52) |
|
| A Short-Term Traffic Flow Prediction Model Based on Temporal Information Enhanced Transformer |
|
ZHANG MingJian1,2, YE BaoLin1,2, DONG Rui2, CHEN Bin2
|
(1.School of In f ormation Science and Engineering, Zhejiang Sci-Tech University Hangzhou 310018, China;
2.Jiaxing Key Laboratory of Smart Transportations, Jiaxing University, Jiaxing 314001, China)
mjzhang998@163.com; yebaolin@zjxu.edu.cn; drui86@zjxu.edu.cn; chenbin@zjxu.edu.cn
|
| Abstract: Traffic flow prediction is one of the key methods to improve the efficiency of road networks and alleviate urban traffic congestion. In order to further enhance the accuracy of traffic flow prediction, a short-term traffic flow prediction model based on temporal information enhanced Transformer is proposed. This model utilizes a multigraph convolutional network to model spatial features from various perspectives, and employs both Long Short-Term Memory (LSTM) networks and Transformers to simultaneously model short-term and long-term temporal features. Additionally, the model adopts a lightweight structure to enhance real-time responsiveness. Compared to baseline methods on the PEMS (Performance Measurement System) dataset, the proposed model shows a 5% improvement in prediction accuracy and a 65% reduction in time complexity. Experimental results demonstrate that the short-term traffic flow prediction model based on temporal information enhanced Transformer effectively extracts spatiotemporal features from traffic flow data while significantly reducing model complexity. |
| Keywords: traffic flow prediction; deep Learning; LSTM; Transformer; GCN |
