Optimising building design with parametric modelling and machine learning

As the realities of climate change unfold, the stakes are higher than ever for engineers and architects. Building designs must now evolve to simultaneously tackle dual challenges: reducing energy consumption and increasing resilience to changing climates. Yet, in the world of design and construction, especially in Central and Eastern Europe, traditional approaches continue to dominate. Expansive glazing, insufficient shading, and high-energy cooling systems still characterise many buildings - approaches conceived in an era before climate change awareness became critical.

Today, it’s time to shift gears and rethink the very foundations of building design to address 21st century environmental demands.

Why does energy optimisation matter more than ever?

The growing frequency of heatwaves and rising temperatures are pushing European buildings to their limits, particularly in office spaces with substantial glazing and resulting internal heat gain vulnerability. Conventional cooling systems, while necessary, consume massive amounts of energy, often counteracting the initial goals of energy efficiency. To make meaningful strides towards carbon neutrality and comply with increasingly stringent EU standards, we need to look beyond simple fixes and instead turn to powerful, data-driven tools that enable effective optimisation.

Harnessing parametric design for optimal performance

Our recent experiment in Poland demonstrates the immense potential of parametric design in reshaping our approach to building energy optimisation. By combining energy modeling software from Integrated Environmental Solutions (IES) with Python-based automation, our team was able to test and simulate a vast range of variables - about 27,000 different combinations, in fact. From the percentage of glazing on the façade to shading types and insulation properties, we examined how each choice influenced the overall energy performance of a building.

Using Design Explorer, an open-source visualisation tool, we quickly filtered and analysed this massive dataset, uncovering optimal configurations tailored to specific conditions. Having the ability to adjust input values to reflect real-world design constraints allowed us to pinpoint the most efficient solutions that could also enhance occupant comfort. And our exploration didn’t end there….

Expanding horizons with machine learning

With such an extensive database, we saw an exciting opportunity: could we leverage this information to create a machine learning model capable of predicting a building's energy efficiency almost instantaneously? This would allow design teams to make informed choices in the early stages without the days of modeling and simulation typically required.

In the next phase, we developed a machine learning model trained on 80% of our data and tested it on the remaining 20%. Although initial accuracy reached nearly 100%, overfitting issues soon became apparent, meaning the model struggled to generalise beyond its training data. Recognising the need for a more balanced and diverse dataset, we scaled our efforts by incorporating data from two additional buildings, reaching over 50,000 observations. This refined model delivered accuracy rates between 70-90% - a promising improvement that demonstrates machine learning's potential to inform real-world design choices.

Building the future with data-driven design

Our experiment underscores the need for collaborative datasets and establishing guidelines for parametric studies. This framework can support AI model development, creating a foundation for advanced tools that will eventually cater to a more diverse range of buildings and climates. Parametric and AI-driven designs hold the potential to tailor solutions specifically to a project’s environment, significantly reducing energy use for both heating and cooling.

By embracing these innovative methods, we’re not only responding to the challenges of climate change, we are also opening doors to a future where buildings are adaptable, energy-efficient, and capable of enhancing occupants' wellbeing. This journey is just the beginning, and as data-driven design gains traction, we’re advancing towards a world where sustainability is at the heart of every structure - empowering us to build a resilient, efficient, and sustainable future for all.