Conference Paper

Proceedings of eSim 2024: 13th Conference of IBPSA-Canada

     

Neural Ordinary Differential Equations for Simulations of Residences Heated with Electric Baseboards

Gabriel Sabbagh, Massimo Cimmino, Benoit Delcroix
Polytechnique Montréal

Abstract: State of the art control schemes for residential HVAC systems are underutilized due to the small margins of improvement compared to the modelling costs. In this study, Neural Ordinary Differential Equations (NODEs) are presented as an approach to address data-driven thermal modelling of residences. These models combine neural networks and differential equations to create continuous dynamic models. To characterise the usefulness of NODEs as a timeseries machine learning technique for residence modelling, their performance is quantified to describe real measurement data. Results show that it is possible to predict the temperatures over 24-hour prediction horizons with an hourly Mean Absolute Error (MAE) of around 0.2 °C.
Keywords: Residence thermal modelling, Data-driven models, Neural Ordinary Differential Equations
Paper: