Technology to trace grid oscillations
Undergraduates at the Korea Institute of Energy Technology (KENTECH) won bronze at the Korea Electric Power Corporation invention and technology innovation competition after developing technology to analyze how changes in electricity consumption at artificial intelligence (AI) data centers affect the power grid.
According to KENTECH on November 6, an energy engineering student team comprising Kim Ji-su, Choi Jae-young and Chae Jae-yeon, supervised by Professor Lee Hyun-woo, entered ModeTrace, a system for tracing the causes of grid oscillations, in the BIXPO 2026 KEPCO invention and technology innovation competition held from November 4 to November 6 at the Kimdaejung Convention Center in Jeonnam-Gwangju.
When low-frequency oscillations occur on the grid, the system checks whether their cause lies within an AI data center and analyzes which workloads affected the grid.
AI data centers use multiple graphics processing units (GPUs) simultaneously to perform large-scale computations. Rapid fluctuations in electricity consumption over short intervals during this process can cause grid oscillations.
Existing monitoring technology analyzes electricity consumption at the data center level, making it difficult to identify the specific computational workloads causing problems.
Individual GPU workloads and corrective measures
The students' system narrows the analysis to individual GPU workloads within a data center. It identifies how much a particular workload contributes to grid oscillations and proposes measures to reduce the problem.
These include staggering the start times of computations across GPUs and scheduling supplementary computations during periods of falling electricity consumption. The measures also include setting a ceiling on electricity consumption.
The system is designed to predict how each measure would change grid conditions and execute the measure once an operator approves it.
The student team used 2 GPUs to apply 3 measures and simulate the grid's response. They found that grid oscillation alerts and assessment results showed a pattern consistent with the advance predictions.
Potential use in grid operations
The technology could help maintain stable grid operations as AI data centers increase in number. It can identify and adjust individual workloads causing problems without restricting electricity use across an entire data center.
Choi said, “The grid environment is changing as inverter-based power sources, including renewable energy, increase,” adding, “We need technology that can proactively address problems at data centers, which consume large amounts of electricity.”
Original Korean article: Maeil Business
English brief by Korea Wind Intelligence, koreawindintel.com
