A Physics-Aware Framework for Short-Term GPU Power Forecasting of AI Data Centers

arXiv cs.AI / 5/7/2026

💬 OpinionDeveloper Stack & InfrastructureModels & Research

共有:

Key Points

The paper presents PI-DLinear, a physics-informed time-series model for short-term (5–80 minute) GPU power forecasting in AI data centers with highly variable workloads.
It models power demand using a multi-node lumped thermal RC network consistent with Newton’s law of cooling, linking power consumption to GPU compute, memory utilization, and temperature via newly derived time-dependent ODEs.
Trained and evaluated on real AI data center data, PI-DLinear delivers more accurate forecasts than existing state-of-the-art models, including transformer-based and non-transformer approaches.
The forecasts are also designed to remain physically consistent during events like power throttling and load transients, not just statistically accurate.
Reported improvements over SOTA span MSE (0.782%–39.08%), MAE (0.993%–51.82%), and RMSE (0.370%–22.28%) across different look-back and prediction windows.

Abstract

AI data centers experience rapid fluctuations in power demand due to the heterogeneity of computational tasks that they have to support. For example, the power profile of inference and training of large language models (LLMs) is quite distinct and big divergences can result in the instability of the underlying electricity grid. In this paper we propose, to the best of our knowledge, the first physics-informed DLinear time-series model that can accurately forecast power utilization of an AI data center 5-80 minutes (short-term forecasting) into the future. The physics, based on a multi-node lumped thermal resistance-capacitance (RC) network consistent with Newton's law of cooling, is captured using newly derived time-dependent ordinary differential equations (ODE) that separately models and interlinks power consumption with the GPU compute and memory utilization and temperature. The resulting model, that we refer to as PI-DLinear, trained and evaluated on a real AI data center dataset and is not only more accurate than the state-of-the-art (SOTA) models tested, but the forecast profile respects the underlying physics under power throttling and load transient events. Relative to the SOTA transformer-based and non-transformer-based models, improvements in forecasting accuracy (averaged across all look-back and prediction windows) range from 0.782%-39.08% for MSE, 0.993%-51.82% for MAE, and 0.370%-22.28% for RMSE.