Dynamic Modeling of AI Data Center Load in PSCAD

The rapid growth of AI data centers has introduced increasingly large and highly dynamic power-electronic loads, posing significant challenges to power system operation and stability. This report presents an open-source electromagnetic transient (EMT) model of a double-conversion uninterruptible power supply (UPS)-based AI data center developed in PSCAD. The model is developed based on the publicly available information and has not been validated against actual hardware, and does not intend to represent any specific data center design. The intent of this model is to provide a conceptual model that captures the phenomena, like ride-through, to understand the origin of these behaviors and develop a basic understanding of mitigation solutions. The model adopts a modular, bottom-up design approach and captures detailed representations of key components, including the rectifier, inverter, bidirectional DC/DC converter, battery energy storage system, variable-frequency drive (VFD) cooling load, and aggregated computing loads. Both switching and average models of the power converters are implemented, with results showing that the average model substantially improves simulation efficiency while preserving the essential low-frequency dynamics. In addition, a centralized voltage and frequency ride-through module is developed to capture UPS protection logic, operating mode transitions, and tripping and reconnection behavior.

The model supports multiple operating mode configurations, including bypass mode, normal converter mode, and battery backup mode, and incorporates realistic operating scenarios such as voltage sag events, fast and slow reconnection strategies, and timevarying computing load profiles. Simulation results validate the capability of the proposed model to accurately represent UPS dynamics and capture critical interactions between large AI data center loads and the power grid.