Simulating the behaviour of existing buildings.
Physics-based building energy simulation tools like EnergyPlus can calculate the energy flows in a building model, subject to assumptions about the weather outside, the building materials and systems, and how occupants use the building. These are commonly used in designing new buildings, but to use these tools to model existing buildings we need to develop calibration processes that match model results to measured data.
Applying data-driven methods to buildings.
Data-driven methods that capitalize on machine-learning and data science have great potential to deliver evidence-based retrofit analysis methods and decision-making tools. Applications include parameter estimation, retrofit targeting, prediction, clustering, benchmarking, calibration and stock modelling. We apply the machine learning libraries developed by big tech companies to solving problems that really matter, for example by taking image recognition algorithms and applying them to time-series data from building thermostats and smart energy meters.
Emulating detailed simulations using machine learning.
Surrogate modelling is emerging as a hybrid approach where machine-learning models are fitted to synthetic data from physics-based models to allow instant exploration of the design space. We developed the Net-Zero Navigator platform to use this concept for early-stage design exploration for new buildings, and the BESOS platform to share our codes with others. There is great potential to apply this to retrofits via model calibration and decision-support tools.