A Thermal Sensation Model for Naturally Ventilated Indoor Environments Based on Deep Learning Algorithms

Abstract

With the emphasis on better sustainability and energy efficiency, natural ventilation has been highly considered in architecture design. Then a thermal sensation model for naturally ventilated indoor environments is critical in ensuring acceptable thermal environments. Since the natural ventilation is dependent on the outdoor environment that changes rapidly, traditional thermal sensation models such as the predicted mean vote (PMV) are not applicable. Considering the nonlinear feature and complexity of the outdoor environment, this study develops a thermal sensation model based on a deep belief neural network (DBN) that could reveal nonlinear patterns in processing big data. The input data are occupant related parameters and indoor & outdoor air parameters, and the output is thermal sensation. This investigation collects data in 10 semi-open classrooms and five laboratories during April and November when natural ventilation is typically used. The accuracy of DBN is verified by comparing its performance with that of three shallow neural networks. For transient and non-uniform naturally ventilated environment, this paper provides an effective solution for quantifying the thermal comfort level.

Keywords: Thermal sensation model, artificial intelligence, Deep belief neural network, Natural Ventilation