TFKAN: Time-Frequency KAN for Long-Term Time Series Forecasting

This repository provides the official PyTorch implementation of our paper:

"TFKAN: Time-Frequency KAN for Long-Term Time Series Forecasting"

TFKAN introduces a dual-branch architecture that integrates Kolmogorov-Arnold Networks (KANs) into both time and frequency domains. By jointly modeling temporal dynamics and spectral characteristics, TFKAN aims to better capture global periodicity and local variations for long-term time series forecasting.

Overview of TFKAN. The proposed framework consists of a time-domain branch and a frequency-domain branch, where KAN-based adaptive representations are jointly learned to enhance long-term forecasting performance.

The paper has been published in Neurocomputing:

📄 Paper: https://www.sciencedirect.com/science/article/pii/S0925231226019569
🔗 DOI: https://doi.org/10.1016/j.neucom.2026.134558

If you find this repository useful in your research, please consider citing our paper.

---

Requirements

Dependencies can be installed using the following command:

pip install -r requirements.txt

---

Getting Started

You can download the datasets as follows:

• Air Quality dataset: UCI Air Quality Dataset
• Other six benchmark datasets: Google Drive Folder

After downloading, place them into the folder:

./dataset/

To train and evaluate TFKAN, simply run:

python run_longExp.py

Alternatively, you can execute a predefined script (e.g., for ETTm1) on a Linux server:

bash ./scripts/ettm1.sh

---

Citation

If you use the code or ideas from this repository, please cite our paper:

@article{KUI2026134558,
title = {TFKAN: Time-frequency KAN for long-term time series forecasting},
journal = {Neurocomputing},
volume = {701},
pages = {134558},
year = {2026},
issn = {0925-2312},
doi = {10.1016/j.neucom.2026.134558},
url = {https://www.sciencedirect.com/science/article/pii/S0925231226019569},
author = {Xiaoyan Kui and Canwei Liu and Qinsong Li and Zhipeng Hu and Yangyang Shi and Weixin Si and Beiji Zou},
keywords = {Time series forecasting, Long-term forecasting, Kolmogorov–Arnold Networks, Frequency domain, Fourier transform}
}

---

Acknowledgements

We appreciate the following GitHub repositories for their valuable codebases and datasets:

1. Informer
2. Autoformer
3. FEDformer
4. FreTS

---