Geographically extended occupant clothing behavior model using convolutional neural networks with hyperband technique

Abstract

Occupant basic clothing insulation is essential in the computation of occupant thermal comfort indices and heating ventilation and air conditioning (HVAC) sizing processes. Consequently, inaccurate estimations of occupant clothing behavior result in poorly designed indoor environments and could indirectly result in unnecessary energy consumption and/or thermal comfort complaints. As a solution, several studies have attempted to develop data-driven predictive models for occupant basic clothing insulation but face two challenges; (I) lack of sufficiently large datasets that are universal and (II) use of simple analytical methods that could potentially misrepresent occupant clothing behavior given its complexity. We develop a convolutional neural network (CNN) based on a large universal dataset, that predicts occupant basic clothing insulation. The developed CNN model accurately predicts occupant basic clothing insulation (R2 = 0.94) and outperforms conventional deep learning architectures. The application of the developed models could substantially reduce clothing-related errors in occupant thermal comfort estimation and HVAC sizing processes.