I spent 22 months working with Professor Maira Kolokotroni for the PRELUDE, an EU Horizon 2020 research project (Grant N° 958345) involving 21 partners across 8 countries. The goal was to make EU housing more energy-efficient through data-driven tools. Based at Brunel University London, my work focused on indoor environmental quality and a “Free Running Model” to keep buildings comfortable while lowering energy use.
On paper, it was a building-science role. In practice, it gave me the foundation I now use in data science and analytics. Here are six takeaways that have stuck with me.
1. Model validation
I built detailed thermal simulations of real buildings and calibrated them against on-site sensor data, using metrics like MAE, RMSE, and CvRMSE (following ASHRAE Guideline 14 and CIBSE TM63).
It’s similar to testing a machine learning model against a holdout dataset. You compare predictions with actual results, measure the error, and record where the physics don’t match. Knowing where a model breaks is what makes it reliable.
2. Avoiding false assumptions in data
To avoid running heavy simulations every time, I built simplified regression models linking outdoor weather to indoor comfort, simple enough for a spreadsheet, but statistically sound for research.
Balancing simplicity with accuracy reinforced a core rule: correlation isn’t causation. It sounds basic, but keeping that top of mind prevents misinterpreting noisy sensor data, survey feedback, or feature importance rankings in ML models.
3. Handling shifting baselines
Part of my role involved creating “morphed” weather files (adjusting rural weather data via the Urban Weather Generator to account for urban heat islands) and projecting them under RCP8.5 climate scenarios.
Underneath the building physics, this was a spatial forecasting problem: keeping models accurate when the baseline environment drifts over time. It’s the exact same challenge data scientists face when deploying models on historical data into shifting markets or conditions.
4. Data ethics with real people
We ran field trials with hostel residents, a vulnerable group requiring strict ethical oversight. That meant designing clear consent workflows, choosing non-intrusive sensors, and returning actionable data to occupants.
Managing informed consent, privacy safeguards, and user-centric study design prepared me directly for responsible AI and GDPR governance.
5. Managing international stakeholders
I led the development of a shared methodology across five pilot buildings in five countries based on European comfort and daylight standards (BS EN 16798-1 and BS EN 17037). Co-writing deliverables with project partners meant constantly translating terminology and priorities between different institutions, building codes, and regional norms.
6. Translating complex technical findings
A major deliverable was turning complex indoor-comfort models into simple, behavioural prompts for residents, like nudging them when to open windows or drop shades.
The skill is similar to writing a clear model card or executive dashboard. A complex model isn’t useful if the end user doesn’t know what action to take.
From urban physics research to data science
I used EnergyPlus, DesignBuilder, and the Urban Weather Generator (UWG) as a powerful research workflow for exploring the links between building energy modelling (BEM), microclimate analysis, and urban physics. But the core skills- rigorous validation, handling uncertainty, ethical data collection, and clear technical communication- translate directly to the analytics and data science work I do today.
Leave a Reply