Channel Adapter for Time Series Foundation Models in Zero-Shot Multivariate Forecasting

Dongyuan Li, Renhe Jiang, Shun Zheng, Zheng Dong, Haotian Gao, Ying Zhang, Jiang Bian

TL;DR: We build a lightweight plug-and-play channel adapter that allows TSFMs to leverage multivariate correlations.

Abstract: Time Series Foundation Models (TSFMs) have achieved strong performance in univariate time series forecasting. However, most TSFMs rely on channel-independent pre-training that models each variable separately, limiting their ability to leverage inter-channel information that is crucial in real-world multivariate systems. Motivated by this limitation, we propose ChaTSFM, a lightweight plug-and-play channel adapter that allows frozen TSFMs to leverage multivariate correlations in a zero-shot setting. ChaTSFM first builds a budgeted pre-training dataset to cover diverse heterogeneous inter-channel dependency patterns. It then uses data-derived domain descriptors to learn a dataset-conditioned inter-channel similarity measure that reduces cross-domain metric distortion. Finally, it injects sparse inter-channel information via gated refinement, leveraging multivariate information without degrading intra-channel temporal dynamics. Extensive experiments on nine benchmarks validate the effectiveness of ChaTSFM, demonstrating consistent zero-shot improvements over four best-performing TSFMs while maintaining scalable deployment. Code is available at Code is available at https://github.com/Clearloveyuan/ChaTSFM.

Lay Summary: Many real-world systems, such as power grids, road networks, and weather stations, are tracked by many sensors whose signals often rise and fall together. When one solar panel dims under a cloud, its neighbors usually do too. Today's powerful AI forecasters can predict such signals well, but they still share a blind spot: they often treat each sensor in isolation, ignoring how sensors influence one another. Yet these relationships are often the information that matters most. We built ChaTSFM, a small add-on that teaches existing forecasters to use sensor relationships without retraining the large model underneath. It learns which sensors genuinely move together and corrects a prediction only when a related sensor provides a useful hint. Across nine real-world datasets, this lightweight add-on consistently improves forecasts with almost no extra computational cost, making accurate forecasting cheaper and more practical for decisions in energy, transportation, and beyond.

Link To Code: https://github.com/Clearloveyuan/ChaTSFM

Primary Area: Applications->Time Series

Keywords: Multivariate Time Series Forecasting, Time Series Foundation Models, Channel Adapter

Submission Number: 2697