Adaptive Multi-Scale Decomposition Framework for Time Series Forecasting

Overview

The Adaptive Multi-Scale Decomposition Framework (AMD) is a cutting-edge solution for time series forecasting, incorporating three main components: the Multi-Scale Decomposable Mixing (MDM) Block, the Dual Dependency Interaction (DDI) Block, and the Adaptive Multi-predictor Synthesis (AMS) Block.

Prerequisites

To get started, ensure you are using Python 3.10. Install the necessary dependencies by running:

pip install -r requirements.txt

Data Preparation

Download the required datasets from Autoformer and iTransfomer. Organize the data in a folder named ./data as follows:

data
├── electricity.csv
├── exchange_rate
├── ETTh1.csv
├── ETTh2.csv
├── ETTm1.csv
├── ETTm2.csv
├── solar_AL.txt
├── traffic.csv
└── weather.csv

Training Example

All training scripts are located in the ./scripts directory. The details of the hyper-parameter settings are in Appendix C.4 in our paper. To train a model using the weather dataset, run the following command:

./scripts/Weather.sh

Citation

If you find this repository helpful, please cite our paper:

@inproceedings{hu2025adaptive,
  title={Adaptive Multi-Scale Decomposition Framework for Time Series Forecasting},
  author={Hu, Yifan and Liu, Peiyuan and Zhu, Peng and Cheng, Dawei and Dai, Tao},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  year={2025}
}

Acknowledgements

We gratefully acknowledge the following GitHub repositories for their valuable contributions:

Contact

For any questions or inquiries, please submit an issue or contact us via email:

Yifan Hu (huyf0122@gmail.com)
Peiyuan Liu (lpy23@mails.tsinghua.edu.cn)

AMD

Install / Use

README