Voltage ride-through for data centers: what the facility does versus what the model claims
A data center rides through a sag in layers, and each layer has a different threshold. The planner is not asking what your UPS tolerates. They are asking what the meter sees.
A data center rides through a voltage sag in layers, and each layer has a different threshold: the service, the UPS, the mechanical plant, and the server power supply all respond differently. Equipment envelopes like ITIC and SEMI F47 describe device tolerance, not facility behavior. The interconnection planner is asking a different question entirely, about the aggregated response at the meter, including how load comes back.
Ride-through happens in layers
When terminal voltage drops, several things in a data center respond on different timescales and with different thresholds. Reasoning about them separately is what keeps the conversation honest.
| Layer | Typical behavior in a sag | What it contributes at the meter |
|---|---|---|
| Server power supply | Holds through short, shallow events using internal energy storage; draws more current as voltage falls while in regulation | Near-constant-power behavior over its regulating range |
| UPS | Depending on topology, transfers to stored energy or continues conditioning; input draw can change sharply | Decouples the critical bus from the grid, and can mask or amplify what is seen upstream |
| Mechanical plant | Chillers, pumps and fans are often not on the critical bus and may trip or stall | A real block of load that can disconnect, and that has its own restart sequence |
| Facility protection | Undervoltage and transfer schemes act on their own settings and timers | The timed cease decision the model has to represent |
The point of the table is not the specific rows, which vary by design. It is that ride-through is not one number. Any single-number answer to “does the facility ride through” is describing one layer and quietly ignoring the others.
Why ITIC and SEMI F47 do not answer the question
Two published envelopes come up constantly in these conversations. Both are useful. Neither is what a transmission planner is asking for.
The ITIC curve, maintained by the Information Technology Industry Council, is a voltage-tolerance envelope for information technology equipment. It describes the sag and swell conditions a device is expected to tolerate. It is written about equipment, and it says nothing about how a facility’s load returns after the event.
SEMI F47 is the analogous sag-immunity specification from the semiconductor industry, aimed at process equipment. Same shape of idea, same limitation for this purpose.
Both describe a floor for device survival. A dynamic model has to describe aggregate behavior through time, including the part that happens after voltage recovers. An envelope cannot express a reconnect threshold, a reconnect delay, or a restoration ramp, and those are often the parameters a study is most sensitive to.
Equipment specifications answer “will this survive?”. Interconnection studies ask “what will the system see, and when does it come back?”. The second question is not derivable from the first.
What the planner is actually asking
Reduced to its core, the interconnection question about ride-through has four parts, and a dynamic model has to take a position on each:
- How much load stays connected as terminal voltage falls, and how its active and reactive draw changes while it does.
- At what point it ceases, and after what timer, if it ceases at all.
- At what voltage it reconnects, and after what delay.
- How quickly it ramps back to pre-event load.
That sequence, ride through, reduce or disconnect, wait, restore, is exactly what the PERC1 structure exists to express, and exactly what a ZIP representation cannot. If you want the model-side detail, the PERC1 guide walks the parameters.
Aggregation changes the answer
A common failure is to reason from one rack to the whole hall. Aggregation is not a multiplication.
Different halls sit at different points in commissioning. Mechanical load is not distributed evenly behind the critical bus. Protection settings differ between blocks, sometimes because the blocks were built in different years by different contractors. The result is that a facility often does not cease as one unit, it ceases in steps, and the steps are what a planner sees.
This cuts both ways for your submission. A single aggregate model with one cease threshold can be conservative and defensible, or it can be wrong in a direction nobody checked. Which one it is depends on whether anybody looked at the block-level settings before choosing the number.
What it takes to prove a ride-through claim
Claims about ride-through split cleanly into two kinds, and they need different evidence.
Protection and reconnect claims can usually be established from settings: relay and transfer-scheme thresholds, timers, and the documented restoration sequence. These are auditable without a disturbance, and they should be locked before any fitting happens, so an optimizer cannot explain a trace by quietly moving a protection threshold.
Continuous-response claims, how active and reactive power actually move with voltage and frequency, need measured behavior. And here the specific disturbance matters more than its severity. A sag that stays above the cease threshold is worth more for calibration than one that trips the load, because a disconnected load reports nothing about its own continuous response. We measured that inversion, along with what decimating a record does to it, in the identifiability note.
Where no suitable record exists, the honest deliverable is the gap and the commissioning test that would close it, not a fitted number that survived review because nobody asked where it came from.
Frequently asked questions
Does the ITIC curve tell the utility how my data center behaves?
No. The ITIC curve is an equipment voltage-tolerance envelope for information technology equipment. It describes what a device should survive, not what your aggregated facility does at the point of interconnection, and it carries no statement about how load returns after a disturbance.
If my UPS rides through a sag, does the grid see anything?
Usually yes. A double-conversion UPS can hold its critical bus while its input draw changes substantially, and mechanical cooling behind it may not ride through at all. What the meter sees is the sum of all of that, which is rarely the same as the tolerance of any one device.
What is the difference between ride-through and reconnect?
Ride-through is whether load stays connected during the disturbance. Reconnect is what happens afterwards: at what voltage it comes back, after what delay, and how fast it ramps. Planners often care more about the second, because simultaneous fast reconnection across a large load can be the bigger system event.