Comfort temperature prediction according to an adaptive approach for educational buildings in tropical climate using artificial neural networks

Abstract

Following the adaptive thermal comfort approach, we show a thermal comfort modeling by artificial neural networks to predict the preferred comfort temperature Tcomf for educational buildings occupants in a tropical climate. The field data collected in 27 educational buildings, climatized with air conditioning systems AC, and natural ventilation NV, in Tuxtla Gutiérrez-México, allow developing the Tcomf modeling by artificial neural network-based models ANN-BM. The predictor variables were the outdoor running mean temperature, relative humidity, air velocity, weight, clothing insulation, and activity level. For AC, the R2 and the mean absolute error were 0.985 and 0.319°C; for NV, 0.981 and 0.286°C. Concerning local models and current standard models, the accuracy improvement to determine Tcomf was significant, following R2; for AC mode, the ANN-BM was 50 times better; for NV was 41.5 times better. The mean Tcomf by ANN-BM was 2.4°C above obtained with the CIBSE guide and 1.3°C concerning a Local model. For NV, the mean Tcomf was 0.5°C above following CEN standard EN15251, 2.0°C above ASHRAE standard 55, and 1.1°C above the Local model. Thus, allowing the air conditioning to run at a higher Tcomf than calculated with current standard models, making it possible to reduce thermal cooling loads while thermal satisfaction also increases.