DUET: Dual Clustering Enhanced Multivariate Time Series Forecasting

This code is the official PyTorch implementation of our KDD'25 paper: DUET: Dual Clustering Enhanced Multivariate Time Series Forecasting.

If you find this project helpful, please don't forget to give it a ⭐ Star to show your support. Thank you!

🚩 News (2024.12) DUET has been included in the time series forecasting benchmark TFB and the time series analytics leaderboard OpenTS.

🚩 News (2024.11) DUET has been accepted by SIGKDD 2025.

Introduction

DUET, which introduces a DUal clustering on the temporal and channel dimensions to Enhance multivariate Time series forecasting. Specifically, it clusters sub-series into fine-grained distributions with the TCM to better model the heterogeneity of temporal patterns. It also utilizes a Channel-Soft-Clustering strategy and captures the relationships among channels with the CCM. Euipped with the dual clustering mechanism, DUET rationally harnesses the spectrum of information from both the temporal and channel dimensions, thus forecasting more accruately.

The important components of DUET: (a) Distribution Router; (b) Linear Pattern Extractor; (c) Learnable Distance Metric; (d) Fusion Module.

Quickstart

1. Requirements

Given a python environment (note: this project is fully tested under python 3.8), install the dependencies with the following command:

pip install -r requirements.txt

2. Data preparation

You can obtained the well pre-processed datasets from Google Drive. Then place the downloaded data under the folder ./dataset.

3. Train and evaluate model

• To see the model structure of DUET, click here.
• We provide all the experiment scripts for DUET and other baselines under the folder ./scripts/multivariate_forecast. For example you can reproduce all the experiment results as the following script:

sh ./scripts/multivariate_forecast/ETTh1_script/DUET.sh

Results

We utilize the Time Series Forecasting Benchmark (TFB) code repository as a unified evaluation framework, providing access to all baseline codes, scripts, and results. Following the settings in TFB, we do not apply the "Drop Last" trick to ensure a fair comparison.

Extensive experiments on 25 real-world datasets from 10 different application domains, demonstrate that DUET achieves state-of-the-art~(SOTA) performance. We show the main results of the 10 commonly-used datasets below, click here to see the results for the remaining 15 datasets:

Citation

If you find this repo useful, please cite our paper.

@inproceedings{qiu2025duet,
 title     = {DUET: Dual Clustering Enhanced Multivariate Time Series Forecasting},
 author    = {Xiangfei Qiu and Xingjian Wu and Yan Lin and Chenjuan Guo and Jilin Hu and Bin Yang},
 booktitle = {SIGKDD},
 year      = {2025}
}

@article{qiu2024tfb,
 title   = {TFB: Towards Comprehensive and Fair Benchmarking of Time Series Forecasting Methods},
 author  = {Xiangfei Qiu and Jilin Hu and Lekui Zhou and Xingjian Wu and Junyang Du and Buang Zhang and Chenjuan Guo and Aoying Zhou and Christian S. Jensen and Zhenli Sheng and Bin Yang},
 journal = {Proc. {VLDB} Endow.},
 volume  = {17},
 number  = {9},
 pages   = {2363--2377},
 year    = {2024}
}

Contact

If you have any questions or suggestions, feel free to contact:

• Xiangfei Qiu (xfqiu@stu.ecnu.edu.cn)
• Xingjian Wu (xjwu@stu.ecnu.edu.cn)

Or describe it in Issues.