DDN: Dual-domain Dynamic Normalization for Non-stationary Time Series Forecasting

This repo is the official Pytorch implementation of our NeurIPS 2024 paper: DDN: Dual-domain Dynamic Normalization for Non-stationary Time Series Forecasting.

Introduction

In this paper, we introduce a novel Dual-domain Dynamic Normalization (DDN) framework designed to address the challenge of dynamically capturing distribution variations across time and frequency domains. DDN operates in a sliding window fashion, enabling it to detect subtle, time-varying changes in data distributions. DDN performs time-domain normalization to compute local sliding statistics (mean and standard deviation) at each time step, offering a fine-grained approach compared to traditional methods that operate at a coarser level.

We perform comparative experiments across seven widely-used datasets using popular forecasting backbones, and further evaluate DDN's performance against other reversible non-stationary approaches.

Multivariate forecasting results:

Comparison with other reversible non-stationary approaches:

Usage

Environment and dataset setup

pip install -r requirements.txt
mkdir datasets

All the datasets are available at the Google Driver provided by Autoformer. Many thanks to their efforts and devotion!

Running

We provide ready-to-use scripts for different backbone models. For example, to run DDN with DLinear:

sh ./scripts/DLinear.sh # scripts for DLinear

Acknowledgement

This repo is built on the pioneer works. We appreciate the following GitHub repos a lot for their valuable code base or datasets:

• FEDformer

• Autoformer

• DLinear

• iTransformer

• SAN

Citation

If you find our work helpful, please consider citing our paper:

@article{dai2024ddn,
  title={{DDN}: Dual-domain Dynamic Normalization for Non-stationary Time Series Forecasting},
  author={Dai, Tao and Wu, Beiliang and Liu, Peiyuan and Li, Naiqi and Xue Yuerong and Shu-Tao Xia and Zexuan Zhu},
  journal={Advances in Neural Information Processing Systems},
  year={2024}
}