Machine Learning Guided Cooling System Optimization for Data Center

TL;DR

A three-stage, physics-guided machine learning framework for identifying and reducing cooling energy waste in high-performance computing facilities yields interpretable recommendations, supports counterfactual analyses such as flow reduction during low-load periods and redistribution of thermal duty across cooling loops, and provides a practical pathway toward quantifiable reductions in accessory power.

Abstract

Effective data center cooling is crucial for reliable operation; however, cooling systems often exhibit inefficiencies that result in excessive energy consumption. This paper presents a three-stage, physics-guided machine learning framework for identifying and reducing cooling energy waste in high-performance computing facilities. Using one year of 10-minute resolution operational data from the Frontier exascale supercomputer, we first train a monotonicity-constrained gradient boosting surrogate that predicts facility accessory power from coolant flow rates, temperatures, and server power. The surrogate achieves a mean absolute error of 0.026 MW and predicts power usage effectiveness within 0.01 of measured values for 98.7% of test samples. In the second stage, the surrogate serves as a physics-consistent baseline to quantify excess cooling energy, revealing approximately 85 MWh of annual inefficiency concentrated in specific months, hours, and operating regimes. The third stage evaluates guardrail-constrained counterfactual adjustments to supply temperature and subloop flows, demonstrating that up to 96% of identified excess can be recovered through small, safe setpoint changes while respecting thermal limits and operational constraints. The framework yields interpretable recommendations, supports counterfactual analyses such as flow reduction during low-load periods and redistribution of thermal duty across cooling loops, and provides a practical pathway toward quantifiable reductions in accessory power. The developed framework is readily compatible with model predictive control and provides a template that, with site-specific recalibration, could be adapted to other liquid-cooled data centers with different configurations and cooling requirements.

Figures (11)

• Figure 1: Flowchart of the physics-guided framework.
• Figure 2: Calibration plot for the accessory-power surrogate on the 2023 test set. Each point is a 10-minute interval; the dashed line denotes perfect prediction, and the solid line with shaded band shows the binned median and 10th--90th percentile of predicted $P_{\mathrm{acc}}$ as a function of the actual value.
• Figure 3: Residual distribution for the accessory-power surrogate on the test set. The histogram and KDE are approximately symmetric and centered near zero, with most residuals confined to a narrow band around the origin.
• Figure 4: Residuals versus predicted accessory power on the test set. Residuals remain centered around zero across the prediction range and show no clear structure or heteroscedasticity.
• Figure 5: Time-series comparison of measured and predicted accessory power on the test set. (a) Full test period. (b) Zoom into a representative high-load week. (c) Zoom into a representative low-load / maintenance week.