Time series models predict power load with uncertain data
Benchmarking Time Series Foundation Models for Load Forecasting Under Covariate Uncertainty
Machine Learning
Summary
Predicting short-term electricity demand helps power systems run smoothly. The authors benchmarked several advanced forecasting models that use past and future information to make predictions. They found one model, Chronos-2, works very well when future information is accurate but struggles when that information is noisy. Another model, TimesNet, is more reliable when the future data is uncertain. This shows that while powerful models exist, handling uncertainty in input data is crucial for real-world power forecasting.
What this means in practice
- •For power grid operators: Choose robust forecasting models to improve power demand predictions when future data quality varies.
- •For energy software developers: Integrate state-of-the-art time series models like Chronos-2 and TimesNet to enhance load forecasting tools under uncertain data conditions.
Authors
Tomas Kaljevic, Ivan Arzola, Yu Zhang
Abstract
Accurate short-term load forecasting (STLF) is essential for the reliable and efficient operation of modern power systems. While time series foundation models (TSFMs) have recently demonstrated remarkable performance across a wide range of forecasting tasks, their effectiveness for STLF under realistic operational conditions remains largely unexplored. In this paper, we present a comprehensive benchmark of four trained-from-scratch (TFS) models and four TSFMs across three real-world load forecasting datasets under operational scenarios that differ in the availability and quality of future covariate information. Our results show that Chronos-2 consistently achieves state-of-the-art performance in both zero-shot and fine-tuned settings when future covariates are available or accurately forecast. However, its performance degrades as covariate forecasts become increasingly noisy, whereas TimesNet exhibits greater robustness under severe covariate uncertainty. These findings demonstrate the effectiveness of covariate-informed TSFMs for STLF while highlighting the critical role of robust covariate modeling in real-world forecasting applications.