Source code for the paper, "Dish-TS: A General Paradigm for Alleviating Distribution Shift in Time Series Forecasting", in AAAI 2023.
Dish-TS is a general paradigm for time series forecasting against distribution shift.
Similar to reversible instance normalization, Dish-TS is model-agnostic such that it can be coupled with any forecasting architectures.
Note that in experiments, we directly take the original data for training/evaluation to directly reflect the distribution shift in time series, and do not use preprocessing techniques (e.g., z-score normalization, min-max normalization) to process time series dataset.
If you find our work interesting, you can the paper as
@inproceedings{fan2023dish,
title={Dish-TS: A General Paradigm for Alleviating Distribution Shift in Time Series Forecasting},
author={Fan, Wei and Wang, Pengyang and Wang, Dongkun and Wang, Dongjie and Zhou, Yuanchun and Fu, Yanjie},
booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
volume={37},
number={6},
pages={7522--7529},
year={2023}
}