Paper Conference

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

     

INDOOR AIR POLLUTION ESTIMATION USING MACHINE LEARNING (ANN AND SVR) IN SMART BUILDINGS

Martin Gabriel, Thomas Auer

DOI: https://doi.org/10.26868/29761662.2022.24
Abstract: This research reports on a case study in an open office, recording indoor air pollution and smart building data in high spatiotemporal resolution for time intervals amounting to more than six months and spread over four seasons. Six measurement nodes recorded indoor air pollution (particulate matter concentration) and smart-building data (temperature, pressure, humidity, sound, illumination, window opening states, and printer power consumption) on a sub-minute time resolution. The data was used to train machine learning models and to evaluate the predictions. Two machine learning typologies are here examined: artificial neural networks (ANNs) and support vector machine regression (SVRs). The models are tested for three individual weeks and evaluated using the R2 and mean average error (MAE) metric. ANNs were found to perform best with maximum R2 of 0.72 and a mean R2 of 0.68 accounting for all weeks.
Paper:
bausim2022_Gabriel_Martin