A thermal sensation model for naturally ventilated indoor environments based on deep learning algorithms

In recent years, with the emphasis on sustainability and energy efficiency, natural ventilation has attracted increasing interest from building designers. Natural ventilation is dependent on the outdoor environments which could change rapidly, and the traditional thermal sensation models such as the predicted mean vote (PMV) are not applicable, correspondingly. The deep belief neural network can reveal nonlinear patterns in processing big data, and it can be used to predict target data with high flexibility and accuracy. This study developed a deep belief neural network model for indoor thermal sensation prediction in naturally ventilated environments with outdoor environment parameters and human factors: outdoor air temperature, average radiant temperature, outdoor air relative humidity, outdoor wind speed, clothing thermal resistance, activity level, gender, age and weight collected in 10 semi-open classrooms and 5 laboratories in April and November when natural ventilation was used. The research compared the performance of deep belief neural networks with three neural networks: BP, Elman and fuzzy neural networks. Results showed that the deep belief neural network can enhance the performance of thermal sensation prediction of natural ventilated indoor environments. The research provides a more flexible and effective solution for thermal comfort prediction of natural ventilated indoor environments.