SDGL- Official PyTorch Implementation

The official implementation of Dynamic graph structure learning for multivariate time series forecasting (paper).

The initial version was Dynamic Graph Learning-Neural Network for Multivariate Time Series Modeling (arxiv).

Requirements

python 3
see requirements.txt

Data Preparation

Raw Dataset

Traffic data https://github.com/LeiBAI/AGCRN

Time series data https://github.com/laiguokun/multivariate-time-series-data

Train Commands

For traffic datasets (PeMSD4, PeMSD8):

python Pems4/train_pems.py --gcn_bool --addaptadj --dataset

For time series datasets:

python Time_series/train_series.py --gcn_bool --addaptadj --dataset

Experimental result

I have placed the prediction results of this model on the website

Citing

If you find this repository useful for your work, please consider citing it as follows:

@article{DBLP:journals/pr/LiZYX23,
  author       = {Zhuo Lin Li and
                  Gao Wei Zhang and
                  Jie Yu and
                  Lingyu Xu},
  title        = {Dynamic graph structure learning for multivariate time series forecasting},
  journal      = {Pattern Recognit.},
  volume       = {138},
  pages        = {109423},
  year         = {2023},
  url          = {https://doi.org/10.1016/j.patcog.2023.109423},
  doi          = {10.1016/j.patcog.2023.109423},
  timestamp    = {Fri, 23 Jun 2023 22:30:47 +0200},
  biburl       = {https://dblp.org/rec/journals/pr/LiZYX23.bib},
  bibsource    = {dblp computer science bibliography, https://dblp.org}
}

ZhuoLinLi-shu / SDGL