Component-based machine learning for energy performance prediction by multiLOD models in the early phases of building design

authored by: Philipp Florian Geyer, Manav Mahan Singh, Sundaravelpandian Singaravel
Abstract: The application of building information modeling (BIM) in early design phases requires the support of different levels of detail (LOD). This allows scaling to be supported as an important activity of designing. Furthermore, to achieve well-performing solutions in terms of energy efficiency, it is necessary to consider energy performance in early design stages. Therefore, this paper presents a multiLOD modeling approach for the early phases of building design that integrates energy performance prediction based on component-based machine learning (ML) using artificial neural networks (ANN). A model structure with three adaptive LOD definitions is proposed to support the design process by a digital model that supports flexible scaling back and forth. By linking the ML models to the elements in this structure, components are formed that support quick and flexible modeling and energy performance prediction in the early building design process. The transformation rules flexibly link the ML components to all LOD. This approach was illustrated and validated by a test case with a medium-sized office building. The early design states of the case were reconstructed for the application of the method. For validation purposes, the results of the ML predictions for 60 different design configurations were compared to those of a conventional parametric full-detail simulation model. This comparison showed that the average error was no higher than 3.8% for heating and 3.5% for cooling.
External Organisation(s): KU Leuven
Type: Conference contribution
Pages: 516-534
No. of pages: 19
Publication date: 19.05.2018
Publication status: Published
Peer reviewed: Yes
ASJC Scopus subject areas: Theoretical Computer Science, General Computer Science
Sustainable Development Goals: SDG 7 - Affordable and Clean Energy
Electronic version(s): https://doi.org/10.1007/978-3-319-91635-4_27 (Access: Closed)