Over the last months, I’ve been working on grid studies for a large data center project in the ERCOT footprint. One of the key tasks was to build a realistic Composite Load Model (CMLD) in PSS®E so that our dynamic and EMT studies actually reflect how the data center behaves during faults, voltage dips and frequency events. In this post I’d like to share how we approached CMLD modeling for the data center, step by step, and some lessons learned that may be useful for others working on similar projects.
Why CMLD matters for data centers in ERCOT
Data centers are not “just a big constant P load”. They are a mix of:
- UPS + IT loads (power electronic, tightly regulated)
- Chiller and cooling plants (large motors)
- Pumps and fans (smaller motors)
- Auxiliaries, lighting and misc. static loads
During a disturbance, each of these behaves differently:
- Motors can stall, trip, or re-accelerate.
- UPS and rectifiers can ride through or disconnect based on DC link and control logic.
- Static loads can be voltage‑dependent (ZIP).
ERCOT increasingly expects realistic load behavior in stability studies, especially for large, concentrated loads connected to weak or IBR‑rich grids. A proper CMLD in PSS®E helps us capture that behavior and avoid overly optimistic results.
Step 1 – Understanding the data center load mix
Our first step was not in PSS®E at all – it was in data gathering.
Together with the client, we built a simplified load breakdown:
- IT/UPS block: share of total MVA, power factor, ride‑through expectations.
- Chillers and big compressors: rated kW, motor type, start characteristics.
- Pumps and fans: approximate motor size distribution.
- Static/auxiliary loads: lighting, building services, misc.
We also defined:
- Normal operating point: MW, Mvar and PF at the high‑side of the data center transformer.
- Expected ramp rates: e.g., load block step changes and maximum MW/min.
- Ride‑through philosophy: under‑voltage and under‑frequency trip levels the owner is targeting.
This gave us the target percentages for each CMLD component: motors, electronic, and static load.
Step 2 – Mapping the physics into the CMLD structure
The PSS®E Composite Load Model gives us:
- A static ZIP load.
- Several motor groups (A, B, C, D) with different inertia and protection characteristics.
- A power electronic / electronic load portion.
We mapped the data center components approximately as:
- Electronic load (P_elec):
- Motors – large (Motor A/B):
- Motors – small (Motor C/D):
- ZIP static load:
For each group, we set:
- Fraction of P and Q at the bus.
- Voltage dependence (ZIP exponents) for the static part.
- Motor parameters: inertia (H), starting voltage, torque–speed curve type, stall and trip settings.
The goal was not to model every single motor, but to capture aggregate behavior that is compatible with ERCOT‑type disturbance scenarios.
Step 3 – Building the model in PSS®E
Once we had the conceptual mapping, we implemented the model in PSS®E:
- Network representation
- Base static load
- CMLD parameterization
- Scenario setup
Step 4 – Tuning and validation
A composite model is only as good as its tuning. We tuned the CMLD in several iterations:
- Sanity checks in load flow
- Dynamic response checks
- Consistency with UPS expectations
In ERCOT‑type studies, the focus is not only on whether the data center survives a single event, but how its behavior impacts system voltage stability, frequency, and interaction with IBRs. A well‑tuned CMLD is essential for that.
Key lessons learned
A few takeaways from this CMLD work for a data center in the ERCOT context:
- Start from engineering reality, not from the model dialog. Talk to the client about their actual equipment, UPS architecture, and protection philosophy before touching parameters.
- Aggregation is your friend. You don’t need to model every motor – you need a credible aggregate split between motors, electronics and static load.
- Ride‑through assumptions must be explicit. Agree in writing what “ride‑through” means (voltage and frequency levels, durations) and reflect that in the CMLD logic.
- Use the same model across studies. Once validated, the same CMLD should be used in steady‑state contingency, dynamic stability, and, when needed, as the basis for EMT scenarios.
- Document everything. For the TSO/ISO (like ERCOT) and for future internal work, clear documentation of CMLD assumptions is just as important as the .RAW/.DYR files.
