Enabling Time-series Foundation Model for Building Energy Forecasting via Contrastive Curriculum Learning

TL;DR

This paper introduces a contrastive curriculum learning approach to adapt foundation models for building energy forecasting, significantly improving zero/few-shot performance in energy time-series prediction tasks.

Contribution

It proposes a novel contrastive curriculum learning method for better adaptation of foundation models to building energy forecasting, addressing limitations of straightforward fine-tuning.

Findings

14.6% improvement in zero/few-shot performance

Effective data ordering enhances model adaptation

New TSFM model available for energy forecasting

Abstract

Advances in time-series forecasting are driving a shift from conventional machine learning models to foundation models (FMs) that are trained with generalized knowledge. However, existing FMs still perform poorly in the energy fields, such as building energy forecasting (BEF). This paper studies the adaptation of FM to BEF tasks. We demonstrate the shortcomings of fine-tuning FM straightforwardly from both the perspectives of FM and the data. To overcome these limitations, we propose a new \textit{contrastive curriculum learning}-based training method. Our method optimizes the ordering of training data in the context of TSFM adaptation. Experiments show that our method can improve the zero/few-shot performance by 14.6\% compared to the existing FMs. Our code and new TSFM will be available at <Anonymous Github Repo>.