Paper Conference

Proceedings of BauSim Conference 2022: 9th Conference of IBPSA-Germany and Austria

Back

---

MACHINE LEARNING FOR BUILDINGS' ENERGY CONSUMPTION PREDICTION IN EARLY DESIGN PHASES

Saskia Elmers, Alexander Hollberg

DOI: https://doi.org/10.26868/29761662.2022.19
Abstract: Reliable prediction of energy consumption is required for energy-efficient design. This paper validates popular models for energy consumption prediction on two public datasets. The models include the mean value, Linear Regression, four Artificial Neural Networks, Support Vector Machine, Random Forests and Gradient Boosting. In these experiments, nonlinear models outperformed linear models. The tree-based models achieve 92% better prediction of electricity consumption than the mean value. Furthermore, the weighting of the input data for the prediction and thus the essential parameters is determined. Based on the essential parameters, consumption prediction and energy-efficient design can be enabled in the early design phases.
Paper:
bausim2022_Elmers_Saskia