In short: Building energy simulation results can vary by 4,000–6,000% for the same building depending solely on which standard occupancy database supplies the people, lighting and equipment data. Peer-reviewed research across six databases, nine building types and three climates quantified the problem and produced a harmonized fix.
Here is an experiment that should worry anyone who has ever made a decision based on a building energy simulation.
Take one building. Model it once, carefully, in EnergyPlus. Hold everything constant geometry, envelope, climate, systems. Change only one thing: the source of the occupancy data, the standardized profiles describing when people are in the building and how much heat, lighting and equipment load they bring with them. Then pick that source from the databases the industry actually uses, the ones embedded in national standards and reference libraries.
A research team including FREDS founders did exactly this, across six widely used occupancy databases, nine building-use types and three climates: Reykjavik, London and Madrid (Rugani, Picco, Salvadori, Fantozzi and Marengo, 2024, published in Energy & Buildings ). The same building, simulated with different “standard” inputs, returned energy consumption results that differed by up to 4000–6000% in the extreme cases, with average differences for each city in the range of 500–1000% within the same building-use category.
Read that again with a practitioner’s eye. Not 5%. Not 50%, the kind of spread engineering judgment might absorb. Five hundred to one thousand percent on average, depending on nothing more than which respectable database the modeller happened to reach for.
These were not obscure sources
The databases compared were not blog downloads. They included profiles from the US National Renewable Energy Laboratory (NREL), the European standard EN 16798-1, the international standard ISO 17772-1, ASHRAE 90.1, and peer-reviewed academic profile sets. Each is a legitimate, citable, professionally defensible choice. A modeler could select any of them, document the choice, pass review and produce an energy figure several times larger or smaller than a colleague who made an equally defensible selection.
The differences come from two distinct layers, and the research separated them deliberately: the absolute load values (how many watts of people, lights and equipment per square meter) and the schedules (the hourly rhythm of how those loads play out across a day and a week). Both varied dramatically between sources. The worst spreads appeared in intensively occupied categories: restaurants, hotels, healthcare where occupancy assumptions do the most work in the energy balance.
Because this matters more than the engine
The simulation industry talks constantly about engine fidelity, which tool calculates radiant exchange more precisely, which handles airflow networks better. This research is a blunt reminder of proportion: those refinements operate at the percent scale, while the inputs feeding them can swing results at the hundreds-of-percent scale. The physics is only as good as the people-data you pour into it.
It is worth being precise about why occupancy does this. People are not shorter term in the energy balance. Occupants determine internal heat gains directly (bodies, lighting, equipment), drive ventilation requirements, and define when systems run at all. In a well-insulated modern building, internal gains can rival the envelope as a driver of demand which is exactly why getting them wrong by a factor of several is catastrophic for the result.
The consequences are not academic:
- The performance gap. The chronic mismatch between predicted and measured building energy use has many causes, but this is a documented, quantified contributor sitting at the front of the pipeline.
- Unstable comparisons. If your simulation and my simulation of competing designs used different profile sources, the comparison may reflect the databases, not the designs.
- Misplaced confidence. A result quoted to one decimal place carries hidden uncertainty orders of magnitude wider than its precision implies unless the input choice is surfaced and defended.
What can actually be done
The paper’s response was not to counsel despair but to engineer a fix: harmonisatio. The team analyzed the six source databases, reconciled their absolute load values into a single coherent set, re-simulated, and measured the residual spread. Harmonizing the load values collapsed the worst-case discrepancies from thousands of percent down towards roughly 90% still material, with the remaining variance driven mostly by the hourly schedules, which the research then addressed as the second harmonization step. The result became a new, unified occupancy database named FREDS in the published paper designed as a robust default for dynamic simulation, especially at early design stages when project-specific occupancy data does not yet exist. (How that database was built, and what sits inside it, is the subject of its own article.)
For practitioners, three working rules fall out of the research:
- Treat the occupancy source as a declared modeling decision, documented alongside weather file and engine version, not a default silently inherited from a template.
- Keep the source consistent across any comparison. Options A and B must share their people-data, or the comparison is meaningless.
- At an early stage, prefer harmonized or evidence-based defaults over which every single standard happens to be closest to hand because at that stage the database is the occupancy model.
The honest conclusion about building energy simulation
There is a temptation to read a finding like this cynically: if inputs can swing results 4000%, why simulate at all? That is the wrong lesson. The right one is that simulation is an instrument, and instruments need calibration discipline. Used with controlled, consistent, harmonized inputs, dynamic simulation remains by far the best available way to understand a building before it exists. The same research program that exposed this problem demonstrates screening models matching detailed ones within 20% when the inputs are governed properly.
The 4000% problem is not a reason to distrust simulation. It is a reason to distrust unexamined defaults and to ask, of any energy figure put in front of you, the one question this research makes unavoidable: whose occupancy data is under this number?
FREDS builds the harmonized database into every simulation by default the input problem, solved at platform level and published in a peer-reviewed journal. See how FREDS handles inputs
Related reading
Frequently Asked Questions
By how much can occupancy database choice affect simulation results?
Which occupancy databases were compared in the research?
Why does occupancy data affect energy simulation so much?
How do you fix the occupancy data problem?
References
Written by Marco Picco, PhD co-founder of FREDS4Buildings and a building-physics lecturer at the University of Lancashire, whose work on building energy simulation has been published in Energy & Buildings , Building and Environment and the IBPSA Building Simulation conference proceedings. About Marco Picco
Reference: Rugani, R., Picco, M., Salvadori, G., Fantozzi, F. and Marengo, M. (2024). A numerical analysis of occupancy profile databases impact on dynamic energy simulation of buildings. Energy & Buildings, 310, 114114. View paper
