Dynamic Modeling of Transmission Networks, Grid-Following Inverters, and UPS-Based AI Data Centers

Abstract

This document presents dynamic models of a transmission network and average models of power-electronic devices and loads in the synchronous $dq$ frame. We model the network in the positive sequence and its synchronous generators with the classical swing equation and a first-order turbine governor. The power-electronic models cover grid-following inverters and AI data centers. Each inverter has a phase-locked loop, a power controller, and a current controller. Each data center connects to the grid through two step-down transformers. Its low-voltage bus feeds an IT load, an induction-motor cooling load, and a constant-impedance load. A double-conversion uninterruptible power supply (UPS) serves the IT load. We model its rectifier, grid-forming output inverter, battery, and bidirectional DC-DC converter. A measured GPU power trace drives the IT demand. We replace every non-smooth operation with a smooth surrogate. This keeps the discretized model continuously differentiable for the nonlinear solver. We also show how we initialize each model at its power-flow operating point.