EMT Modeling and Coordinated Mode-Transition Ride-Through Control of Modern Data Centers

Overview

The authors build a PSCAD/EMTDC electromagnetic-transient model of a 100 MW data centre (80 percent IT, 20 percent motor cooling load). It includes a UPS with battery, directly connected chiller and cooling-tower induction motors, and a VFD-driven CRAH that can be moved onto UPS power. The work is motivated by a July 2024 event in which about 1,500 MW of data centre load moved to backup power at once.

A voltage-based four-state machine moves the UPS between eco mode (bypass), double conversion, battery, and a ramped recovery in which the rectifier reconnects gradually while the battery carries the difference. The normal band is 0.9-1.15 p.u.; low-voltage cease (go to battery) is at 0.7 p.u. after 20 ms, high-voltage cease at 1.15 p.u. after 500 ms, and the rectifier ramps to rated power in 1 s.

In eco mode the server load and CRAH are fed directly from the grid with the UPS in bypass and the rectifier kept energised to hold DC-link and battery charge; short sags inside the tolerance envelope pass straight through to the server power supplies. During recovery, if another sag occurs the ramp is aborted and the system returns to battery mode, which avoids repeated full-power reconnection during auto-reclosing.

The model is benchmarked against the PSS/E PERC1 and CMPLDWG aggregate models for a three-phase fault and LVRT/HVRT events and agrees well, while also capturing motor re-acceleration and converter transients. In the LVRT case the motor load overshot to about 50 MW (PSCAD) or 55 MW (PSS/E) on re-acceleration from a 20 MW pre-fault level, and POI power recovered to about 95 MW from about 100 MW. The document is a preprint, not a peer-reviewed publication.

Key points

1. In July 2024 a 230 kV line fault in the Eastern Interconnection caused about 1,500 MW of data centre load to transfer to backup power at the same time, and system voltages rose above normal limits.
2. In eco mode the server load and CRAH are fed from the grid with the UPS in bypass and the rectifier energised; short sags inside the tolerance envelope pass to the server power supplies.
3. Mode-transition settings: normal band 0.9-1.15 p.u.; battery at 0.7 p.u. after 20 ms; high-voltage cease 1.15 p.u. after 500 ms; rectifier ramp 1 s.
4. During recovery the battery supplies the gap between IT demand and the ramping rectifier; the ramp is aborted if another sag occurs.
5. Modelled facility: 100 MW with 80 percent IT load and 20 percent motor load.
6. In the LVRT case motor load overshot to about 50 MW (PSCAD) or 55 MW (PSS/E) from 20 MW; POI power recovered to about 95 MW from about 100 MW.

Using it in design

• Store operating-mode behaviour in the UPS catalogue and electrical model: eco-mode availability, transfer thresholds and timers (for example 0.7 p.u./20 ms to battery), and a post-event rectifier ramp time.
• Show the efficiency gain of eco mode alongside its ride-through exposure, since short sags reach the IT power supplies.
• Check that battery autonomy covers the recovery ramp and repeated reclosing events.
• Flag directly connected cooling motors as a re-acceleration inrush risk for generator and grid studies.

Who should read it

Electrical engineers working on UPS and large-load interconnection studies, and developers dealing with grid operators.

Related standards

• IEC 62040-3
• ITIC/CBEMA curve
• IEEE 2800

Publication details

Authors

Junyuan Zheng; Salish Maharjan; Min Lwin; Soummya Roy; Zhaoyu Wang; Wenlong Shi; Feng Qiu; Rajarshi Roychowdhury

Publisher

arXiv (preprint, submitted to IEEE)

Year

2026

Document number

arXiv:2611.08614

Document language

English

Source type

Scientific

Access

Free to read

DOI

10.48550/arXiv.2611.08614

Licence

arXiv distribution permission — reuse conditions remain with the rights holder

Citation