Building Energy Efficiency Estimations with Random Forest for Single and Multi-Zones

Conference Paper

2024

Published in: https://papers.cumincad.org/cgi-bin/works/paper/ecaade2024_167

Surrogate models (SM) present an opportunity for rapid assessment of a building's performance, surpassing the pace of simulation-based methods. Setting up a simulation for a single concept involves defining numerous parameters, disrupting the architect's creative flow due to extended simulation run times. Therefore, this research explores integrating building energy analysis with advanced machine learning techniques to predict heating and cooling loads (KWh/m2) for single and multi-zones in buildings. To generate the dataset, the study adopts a parametric generative workflow, building upon Chou and Bui's (2014) methodology. This dataset encompasses multiple building forms, each with unique topological connections and attributes, ensuring a thorough analysis across varied building scenarios. These scenarios undergo thermal simulation to generate data for machine learning analysis. The study primarily utilizes Random Forest (RF) as a new technique to estimate the heating and cooling loads in buildings, a critical factor in building energy efficiency. Following that, A random search approach is utilized to optimize the hyperparameters, enhancing the robustness and accuracy of the machine learning models employed later in the research. The RF algorithms demonstrate high performance in predicting heating and cooling loads (KWh/m2), contributing to enhanced building energy efficiency. The study underscores the potential of machine learning in optimizing building designs for energy efficiency.

• Publisher Name: eCAADe