Poster Publication in BuildSys ’20

A new poster paper has been published in the Proceedings of the 7th ACM International Conference on Systems for Energy-Efficient Buildings, Cities, and Transportationas as part of the ongoing project by the University of Applied Sciences in Munich.

The authors of the paper are Manuel Weber (Munich University of Applied Sciences), Dr. Christoph Doblander (Technical University of Munich), and Prof. Dr. Peter Mandl (Munich University of Applied Sciences). The published paper is available on the proceedings' website. Moreover, a longer detailed version is available on arXiV.

Paper Abstract

Information about room-level occupancy is crucial to many building-related tasks, such as building automation or energy performance simulation. Carbon dioxide levels and other indoor environmental factors can be used as a proxy to detect occupancy. In this regard, machine learning solutions have been proposed, with solid performance in detecting presence, as well as counting the number of present occupants, if enough training data is available. The challenge is, to collect sufficient room-specific ground truth data for model training. With this poster, we address the use of knowledge transfer from synthetic data to reduce the amount of required real world data. We outline two approaches for the combination of transfer learning with physical simulations, and motivate the generation of additional synthetic data. Our results show that the required real world training data can be reduced by 50%.

Citation

To cite the paper, please use the following BibTex entry:


@inproceedings{
10.1145/3408308.3431124,
author = {Weber, Manuel and Doblander, Christoph and Mandl, Peter},
title = {Detecting Building Occupancy with Synthetic Environmental Data},
year = {2020},
isbn = {9781450380614},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3408308.3431124},
doi = {10.1145/3408308.3431124},
abstract = {Information about room-level occupancy is crucial to many building-related tasks, such as building automation or energy performance simulation. Carbon dioxide levels and other indoor environmental factors can be used as a proxy to detect occupancy. In this regard, machine learning solutions have been proposed, with solid performance in detecting presence, as well as counting the number of present occupants, if enough training data is available. The challenge is, to collect sufficient room-specific ground truth data for model training. With this poster, we address the use of knowledge transfer from synthetic data to reduce the amount of required real world data. We outline two approaches for the combination of transfer learning with physical simulations, and motivate the generation of additional synthetic data. Our results show that the required real world training data can be reduced by 50%.},
booktitle = {Proceedings of the 7th ACM International Conference on Systems for Energy-Efficient Buildings, Cities, and Transportation},
pages = {324–325},
numpages = {2},
keywords = {deep neural network, Synthetic data, transfer learning, CO2},
location = {Virtual Event, Japan},
series = {BuildSys '20}
}

This research was supported by IBM providing free access to IBM® Power System™ AC922 & LC922.

Posted by Simon Fuchs