Skip to main content

Load-aware supercapacitor-based grid-forming energy storage for mitigating AI data center impacts on grid frequency

Resource type
Thesis type
(Thesis) M.A.Sc.
Date created
2026-07-20
Authors/Contributors
Abstract
Artificial intelligence data centers (AIDCs) can introduce fast, repetitive megawatt-level load variations as training workloads alternate between compute and communication phases. These rapid transitions can adversely affect grid frequency and produce sustained frequency oscillations, particularly in low-inertia power systems. This thesis proposes a supercapacitor energy storage system (SESS) interfaced through a grid-forming inverter (GFMI) and controlled using a load-aware virtual synchronous generator (VSG) strategy at the AIDC point of common coupling (PCC). The controller separates the measured AIDC demand into a moving-average component supplied by the upstream grid and a fast-varying component assigned to the SESS through the VSG active-power reference. Short-duration active-power pulse terms are additionally added to the VSG reference power during detected load transitions to improve the initial transient response. The proposed controller is evaluated using MATLAB/Simulink electromagnetic transient simulations (EMT) and compared with grid-only operation, droop-GFMI, and conventional VSG-GFMI control. Under repetitive AIDC training-load cycles, the load-aware VSG-GFMI reduces the worst downward rate of change of frequency (RoCoF) magnitude from 0.45 Hz/s to 0.18 Hz/s, improves the worst-case frequency nadir from 59.79 Hz to 59.94 Hz, and reduces the peak-to-peak frequency envelope from 0.41 Hz to 0.096 Hz. The RMS oscillatory component of upstream-grid power is also reduced from 0.4325 MW to 0.2347 MW. These results demonstrate that load-aware VSG control can use local SESS support to reduce the fast AIDC disturbance transferred to the upstream grid while maintaining support capability under repetitive operation
File
Extent
116 pages.
Identifier
etd24468
Copyright statement
Copyright is held by the author(s).
Permissions
This thesis may be printed or downloaded for non-commercial research and scholarly purposes.
Academic Supervisor
Thesis advisor: Wang, Jason Jiacheng
Language
English
Download file Size
etd24468.pdf 5.84 MB