Research Article
A Short Term Traffic Prediction Model Based on Deep Capture of Temporal Periodic Drift
@INPROCEEDINGS{10.1007/978-3-031-70507-6_15, author={Yong Liu and Jianxun Cui and Zhaohua Long}, title={A Short Term Traffic Prediction Model Based on Deep Capture of Temporal Periodic Drift}, proceedings={IoT as a Service. 9th EAI International Conference, IoTaaS 2023, Nanjing, China, October 27-29, 2023, Proceedings}, proceedings_a={IOTAAS}, year={2024}, month={10}, keywords={Short-term traffic flow prediction Temporal periodic drift Attention mechanism}, doi={10.1007/978-3-031-70507-6_15} }- Yong Liu
Jianxun Cui
Zhaohua Long
Year: 2024
A Short Term Traffic Prediction Model Based on Deep Capture of Temporal Periodic Drift
IOTAAS
Springer
DOI: 10.1007/978-3-031-70507-6_15
Abstract
Accurate prediction of short-term traffic flow is crucial for the control and guidance of urban traffic. This paper proposes a new deep learning traffic flow prediction model, Spatial-Temporal fusion model based on Deformable Convolution (DConv-ST), which deeply captures the spatiotemporal correlations present in the sequences. The model divides the raw data into three types of sequences containing information about recent, daily, and weekly periods. A deformable convolutional module is constructed to solve the problem of temporal periodic drift in the sequence, and graph attention network and multi-head attention mechanism are used to capture local and global spatial correlations. A gated mechanism is used to fuse the results of each component module for output. Experiments including model performance analysis, important component analysis, and ablation analysis were conducted on the publicly available transportation network datasets PeMS04 and PeMS08. The experimental results all demonstrate the superior performance of the proposed model.
