Official implementation of ST-CSL — Spatio-temporal fusion and contrastive learning for urban flow prediction (Knowledge-Based Systems 2023). ST-CSL fuses temporal-view and spatial-view contrastive learning so an urban flow predictor captures global periodicity and the hidden flow relationships between functionally similar regions.
📄 Paper: https://doi.org/10.1016/j.knosys.2023.111104 · 🌐 Project page with abstract, FAQ and BibTeX: https://codezx6.github.io/papers/st-csl.html · 👤 Author: Xu Zhang
A deep learning framework for urban flow prediction leveraging contrastive self-supervised pretraining and multi-component spatio-temporal modeling.
ST-CSL addresses the challenge of spatio-temporal flow prediction in urban environments through a novel contrastive learning framework that captures temporal closeness, period, and trend dependencies.
- Multi-component Architecture: Separate encoders for closeness, period, and trend patterns
- Contrastive Pretraining: Self-supervised representation learning through spatial contrastive objectives
- Residual Architecture: Deep residual networks for robust feature extraction
The ST-CSL framework consists of:
- Component Encoders: Process closeness, period, and trend dependencies independently
- Contrastive Module: Learns spatial representations through contrastive objectives
- Fusion Network: Aggregates multi-component features for final prediction
If you use this code in your research, please cite:
@article{zhang2023stcsl,
title = {Spatio-temporal fusion and contrastive learning for urban flow prediction},
author = {Zhang, Xu and Gong, Yongshun and Zhang, Chengqi and Wu, Xiaoming and Guo, Ying and Lu, Wenpeng and Zhao, Long and Dong, Xiangjun},
journal = {Knowledge-Based Systems},
year = {2023},
volume = {282},
pages = {111104},
doi = {10.1016/j.knosys.2023.111104},
issn = {0950-7051},
url = {https://doi.org/10.1016/j.knosys.2023.111104}
}This project is licensed under the MIT License.