Building Simulation, Date: 2015/01/01 - 2015/01/01
Publication date:
2015-01-01
Pages:
106 -
113
ISSN:
978-93-5230-118-8
14th Conference of International Building Performance Simulation Association, Hyderabad, India, Dec. 7-9, 2015
Author:
Maderspacher, Johannes
Geyer, Philipp ; Auer, Thomas ; Lang, Werner
Abstract:
If detailed building models are applied for long- Term simulations, for instance the prediction of the future energy demand under climate change, the computational effort can turn into a serious issue. Machine learning algorithms like Neural Networks (NN) or Support Vector Machine (SVM) could be an alternative. In this work a possible application of NN and SVM for long- Term forecasts are proven and their limitations are presented. In the examined case study, with a simulation period over 30 years, the SVM is hundred fifty times and the NN ten times faster than a detailed building model. This reduction of computational effort can be useful for further studies as a uncertainty analysis of climate change.