This repository contains the PyTorch implementation of our ICLR'25 paper, “Air Quality Prediction with Physics-Guided Dual Neural ODEs in Open Systems.” In this work, we introduce Air-DualODE for predicting air quality at both city and national levels. Our model is composed of three key components: Physics Dynamics, Data-Driven Dynamics, and Dynamics Fusion.
🚩 News (2025.1) Air-DualODE has been accepted by ICLR 2025 (poster).
- python >= 3.9
pip install -r requirements.txtBeijing: https://www.biendata.xyz/competition/kdd_2018/
KnowAir: https://github.com/shuowang-ai/PM2.5-GNN
cd RunBeijing
python train.py --config_filename ../Model_Config/Beijing/Air-DualODE_config.yaml --des 1
python eval.py --config_filename ../Model_Config/Beijing/Air-DualODE_config.yaml --des 1KnowAir
python train.py --config_filename ../Model_Config/KnowAir/Air-DualODE_config.yaml --des 1
python eval.py --config_filename ../Model_Config/KnowAir/Air-DualODE_config.yaml --des 1If you find this repo useful, please cite our paper.
@article{tian2024air-dualode,
title={Air quality prediction with Physics-Guided dual neural odes in open systems},
author={Tian, Jindong and Liang, Yuxuan and Xu, Ronghui and Chen, Peng and Guo, Chenjuan and Zhou, Aoying and Pan, Lujia and Rao, Zhongwen and Yang, Bin},
journal={ICLR},
year={2025}
}