Energy-efficient buildings rarely perform the way they’re designed to. You might design an apartment with great natural ventilation and daylighting, but it still ends up using far more heating and cooling than planned simply because residents open windows at the wrong times or leave blinds up when they shouldn’t. Most people don’t have a background in building physics, so they can’t tell whether opening a window right now will cool the place down or trap a wave of heat.
- Article title: Predicting indoor environmental conditions using correlation models for behaviour change suggestions
- Published via: Building Services Engineering Research & Technology, (2025);46(6):753-774
- Project: PRELUDE, funded by the European Union’s Horizon 2020, Grant Agreement N° 958345.
- Authors: May Zune, Thet Paing Tun, Tristan de Kerchove d’Exaerde and Maria Kolokotroni
Translating weather forecasts into resident actions
We think there’s a low-tech fix for this: figure out how outdoor weather affects a specific room, then tell residents in plain terms what tomorrow’s forecast means for them. If you borrow from Fogg’s Behaviour Model, people change habits when ability, opportunity, and motivation line up. We focused on ability. Link a room’s conditions closely enough to the weather outside, and tomorrow’s forecast becomes a simple instruction rather than a science lesson.
To see if this actually works, we tested the approach on a lived-in one-bedroom apartment on the 8th floor of a 56-unit building in Geneva. After building a thermal simulation and validating it against real room temperatures from April to June 2023, we ran a full year of hourly simulations across three everyday scenarios: windows kept shut, windows opened on a fixed schedule, and overnight “night purge” cooling.
By mapping how temperature, air change rate, and daylight related to outdoor weather in each scenario, we got a set of straightforward equations. Plug in tomorrow’s forecasted temperature and sunlight, and out comes a prediction for indoor conditions.
Key simulation findings
A few clear patterns emerged:
- Temperature tracking: Indoor and outdoor temperatures were tightly linked, but only within each scenario’s bounds. A heating-schedule model and a free-floating model need completely different equations, and neither holds up outside the weather range it was calibrated for.
- Daylight surprises: Daylight levels were best predicted hour by hour because morning and afternoon sun hit rooms differently. Indoor daylight sometimes crossed the overheating threshold (500 lux) while temperature predictions still looked comfortable. Shading needs its own approach because daylight and thermal comfort don’t always move in the same direction.
- Air quality limits: Closed windows kept infiltration around 0.7 air changes per hour, while opening them bumped that to 2–5. When the apartment was fully occupied,
routinely crept above our 900 ppm benchmark unless ventilation was timed deliberately.
From Excel spreadsheet to daily app action
Right now, the whole system only runs in Excel. In the future, we want to turn it into something an engineer or building manager can set up and use without needing any machine-learning expertise or specialist skills. But a spreadsheet won’t change daily habits by itself. The goal is to feed these outputs into a lightweight mobile dashboard that gives a resident a one-second instruction: “Open your window for two hours this evening,” rather than a graph they have to decode.
A scalable approach for real-world buildings
The specific equations we found are tuned to one Swiss apartment and won’t transfer to another building as-is. But the core workflow is what matters: a tool simple enough for a practitioner to build, accurate enough to guide real behaviour, and honest about where its math stops being reliable.
Leave a Reply