What is PERC1? A practical guide to the data-center load model
PERC1 gives planners a positive-sequence representation of how aggregated power-electronic load responds, ceases, reconnects, and ramps. The hard part is not obtaining the model—it is supporting its parameters with facility evidence.
PERC1—the Power Electronic Reconnecting and Ceasing load model—is a positive-sequence aggregate model for data centers and other concentrations of power-electronic load. It represents active and reactive power response, current limits, voltage/frequency sensitivity, cease logic, reconnection, and restoration. A PERC1 file is useful only when its site-specific parameters can be traced to settings, records, or an explicitly stated assumption.
What PERC1 means
PERC1 stands for Power Electronic Reconnecting and Ceasing. PowerWorld’s official model documentation describes it as appropriate for an aggregation of many power-electronic loads, explicitly including a large number of computers in a data center. The model is built for the bulk-system timescale: it describes what the aggregated load does at its electrical terminal during a disturbance, not the switching action of each server power supply or UPS module.
That distinction matters. A transmission planner does not need a circuit-level replica of every rack to ask whether 300 MW will remain connected through a voltage sag, cease in one or more blocks, or return too quickly after voltage recovers. PERC1 provides a compact language for those positive-sequence questions.
Primary source: PowerWorld, “Load Characteristic Model: PERC1”.
Why a ZIP load model falls short for a data center
A conventional ZIP model divides load into constant impedance, constant current, and constant power fractions. That is useful for steady-state and voltage-dependent behavior, but the usual ZIP representation has no timed cease decision, reconnect threshold, reconnect delay, or restoration ramp. It cannot express the sequence that often matters most in a computational-load event: ride through, reduce or disconnect, wait, and restore.
This is not a theoretical edge case. NERC’s March 2026 aggregated response covered 33,282 MW of operating data-center load. It reported that 25,504 MW—roughly three quarters—relied on the inherent ZIP representation because the submitted protection-device model was blank or “None.” NERC characterized roughly three quarters of the models as insufficient to represent data-center dynamic behavior.
A model that holds power constant is not evidence that the facility will hold power constant.
Primary source: NERC Level 2 aggregated report, March 2026.
How the PERC1 model behaves
It helps to read PERC1 as four connected parts rather than a long parameter row.
- Continuous response. Voltage and frequency deviations drive active and reactive power through gains, filters, lead-lag blocks, exponents, deadbands, and current limits.
- Cease logic. Voltage conditions can start timers that remove a defined fraction of the electronic load. A short excursion may reset the timer; a sustained one may change the fraction online.
- Reconnect logic. Recovery thresholds and delays determine when eligible load begins to return. Reconnect is not simply the reverse of cease.
- Restoration. The returning fraction can be limited and ramped rather than appearing instantaneously. This is what turns a protection event into a system-recovery question.
The active- and reactive-power paths have distinct parameters. The model also distinguishes low and high frequency deadbands, and it can return only part of the ceased fraction after a reconnect. Exact initialization and parameter semantics depend on the implementation and simulator version, which is why a submission should name the target tool and version rather than promise abstract “PERC1 compatibility.”
PERC1 vs. ZIP, CMLD, and a user-defined model
The useful question is not which acronym wins. It is which representation covers each part of the facility and each study question.
| Model | Best use | What it captures | Where it stops |
|---|---|---|---|
| ZIP | Static voltage dependence and baseline studies | Constant Z, I, and P fractions | No timed cease, reconnect, or restoration sequence |
| CMLD | Mixed facility composition | Distribution equivalent plus motor, electronic, and static components | Still needs site-specific composition and component parameters |
| PERC1 | Aggregated compute/UPS power-electronic behavior | P/Q dynamics, limits, cease, reconnect, fraction online, and ramp | Not detailed EMT behavior or proof of internal device modes |
| UDM / OEM | Behavior a standard aggregate cannot represent | Project- or vendor-specific controls and logic | May create portability, disclosure, licensing, and review burdens |
A data center is rarely 100 percent compute load. Cooling motors, drives, lighting, transformers, capacitors, backup generation, and BESS behavior may need other components. Dominion Energy Virginia’s current criteria allow a CMLD, PERC1, user-defined model, or a combination so the mechanical and compute portions can be represented separately. That is a composition decision, not a reason to put the entire site behind one default PERC1 block.
What data is needed to parameterize PERC1?
NERC’s May 2026 Level 3 alert tells transmission planners and planning coordinators to use PERC1, or a model with equivalent or better capabilities, as a baseline for computational-load data requirements. The alert’s requested information is a practical starting point for a project evidence register:
- minimum and maximum MW, expected power factor, seasonal values, and the buildout schedule;
- UPS settings and configuration that affect dynamic performance;
- IT versus non-IT composition at relevant loading levels;
- expected down-ramp and up-ramp rates;
- voltage and frequency protection that can isolate or reduce facility demand;
- reconnect voltage, frequency, timing, method, and restored fraction;
- onsite generation, BESS, islanding, transfer, and demand-response behavior; and
- facility use, including whether the compute is training, inference, storage, or another duty.
The NERC alert is not itself a Reliability Standard and says it does not create a mandatory obligation to implement its Essential Actions. The governing requirement for a project comes from the applicable authority and current agreement. Treat every checklist as versioned source data.
Primary source: NERC Level 3 computational-load alert, May 2026.
How to validate a PERC1 data-center load model
Parameterization and validation are separate jobs. A settings sheet may establish that a voltage threshold is configured at a particular value; it does not prove the site-level fraction that will cease or the aggregate ramp after reconnect. A disturbance trace may reveal the response, but only if it contains the voltage or frequency excitation needed to distinguish the parameters.
- Name the target. Record the authority, study stage, simulator, simulator version, model structure, and electrical bases before fitting anything.
- Build provenance. Preserve the source file, channel mapping, units, time base, transformations, and hash for every setting and terminal record.
- Lock discrete settings. Use reviewed protection and reconnect settings as evidence-backed constraints. Do not let an optimizer move a threshold simply because it improves one trace.
- Screen excitation. Determine which parameters the record can identify. A flat frequency channel cannot support a frequency-droop claim.
- Fit the allowed continuous parameters. Use bounded, physically defensible ranges and retain both the candidate and the starting point.
- Score a real holdout. Prefer a different disturbance class, not only another event with the same shape.
- Replay the exported artifact. Run the actual PSS®E or PowerWorld handoff in the named version and compare its trace with the reference implementation.
- Package the refusal. For every unsupported claim, state what is missing and specify the commissioning test that would supply it.
For a worked explanation of the excitation problem, see our technical note on what a data-center dynamic model can be identified from. For the end-to-end accepted-tool workflow, continue with PSS®E dynamic model validation using disturbance data.
What a PERC1 model does not prove
PERC1 can be the right model class and still be the wrong facility model. A clean import does not prove the parameters. A low residual on one event does not establish that the fit is unique. A positive-sequence match does not establish EMT behavior, harmonics, firmware logic, or every transfer mode behind the point of interconnection.
The defensible claim is narrower and more useful: for the named study class, tool version, operating point, and disturbance envelope, the aggregate terminal response is supported by these settings and records. Everything outside that boundary should be labeled as an assumption, a manual review, or a test still to run.
Frequently asked questions
What does PERC1 stand for?
PERC1 stands for Power Electronic Reconnecting and Ceasing. It is an aggregate positive-sequence load model for groups of power-electronic loads.
Is PERC1 required for every data center?
No universal rule requires PERC1 for every project. NERC’s May 2026 alert says planners should use PERC1 or a model with equivalent or better capabilities as a baseline for computational-load requirements, while the applicable transmission owner or planning coordinator sets the project-specific requirement.
Can PERC1 replace an EMT model?
No. PERC1 is a positive-sequence aggregate model. It can answer many bulk-system planning questions, but it does not reproduce switching-frequency behavior, detailed controls, harmonics, or every internal transfer mode that an EMT or OEM model may be needed to study.
Can a PERC1 model be built from settings alone?
Settings can establish protection and reconnect logic, but measured or commissioned behavior is still needed to support continuous-response parameters and validate the aggregate response. Where no suitable record exists, the honest output is a test plan rather than guessed parameters.