Mohamed Shamseldein, a researcher at the University of Michigan, has proposed a novel control framework for centralized medium-voltage uninterruptible power supply (UPS) systems in large data centers. This research, published in the IEEE Transactions on Power Electronics, aims to transform data centers from potential grid liabilities into assets by managing their power draw more effectively.
Large data centers, particularly those running AI workloads, can exhibit highly dynamic power consumption patterns. This can stress weak points in the electrical grid, especially during faults or when workloads pulse rapidly. Shamseldein’s proposed control architecture operates in three modes to address these challenges. In Mode 1, it regulates a DC stiff bus and shapes the data center’s normal operation grid draw. Mode 2 enforces current-limited fault-mode active and reactive power (P-Q) priority, utilizing the UPS battery energy storage system (UPS-BESS) for buffering and a rate-limited post-fault “soft return.” Mode 3 optionally provides droop-based fast frequency response via grid-draw modulation.
The research demonstrates the effectiveness of this control framework through simulations. In a 50 MW data center scenario with a short-circuit ratio (SCR) of 1.5 and a 150 ms three-phase voltage dip, the proposed framework ensured zero unserved information-technology (IT) energy, reduced peak inverter current, maintained a nonzero mean fault-window grid draw, and improved the settled point-of-common-coupling (PCC) voltage minimum. Additionally, a forced-oscillation case study showed that the normal-operation shaping filters the oscillation seen by the grid while the UPS-BESS buffers the pulsing component.
For the energy sector, this research offers a practical approach to integrate large, dynamic data center loads into the grid more effectively. By transforming data centers into assets that can support grid stability, this control framework could help accommodate the increasing power demands of AI workloads while enhancing overall grid resilience. The research was published in the IEEE Transactions on Power Electronics, a peer-reviewed journal dedicated to the development and practical application of power electronics technology.
This article is based on research available at arXiv.