Traffic prediction is a complex spatiotemporal forecasting task, affected by high dimensional multi factors including fixed road structures and time variant external incidents. The prediction faces the challenges of spatial dependencies of road networks [1] affected by historic traffic data and complex external factors such as the uncertain weather variations and the Point of Interest (POI) attributes attached to the road networks. To address these challenges, we propose a spatiotemporal multi-factor network (STMFN) which is a deep learning framework to model traffic prediction with Graph Convolutional Network (GCN) incorporating a traffic block and a weather block. Specifically, in the traffic block, external factors are regarded as traffic intrinsic attributes to extract spatiotemporal features for predicting future traffic conditions. And the weather block uses the cross-attention mechanism to extract the relationship between weather and traffic flow data. Comprehensive experiments on public benchmark datasets demonstrate the effectiveness of the proposed model.
Spatio-Temporal Multi-Factor Network Based on Attention Mechanism for Traffic Prediction
2024-12-09
509284 byte
Conference paper
Electronic Resource
English
Spatio-temporal graph attention networks for traffic prediction
Taylor & Francis Verlag | 2024
|Traffic Flow Forecasting of Graph Convolutional Network Based on Spatio-Temporal Attention Mechanism
Springer Verlag | 2023
|Track Prediction Based on Spatio-Temporal Attention
Springer Verlag | 2022
|Spatio-Temporal Graph Attention Convolution Network for Traffic Flow Forecasting
Transportation Research Record | 2024
|