| description abstract | In the current work, the artificial intelligence techniques multilayer perceptron neural network (MLP), radial basis function neural network (RBF), and group method of data handling (GMDH) were used to estimate the indoor temperature (Tint) in buildings under tropical climate conditions. The data used for modeling correspond to experimental samples measured during one year, considering as a case study the installation of a university laboratory. The models were developed utilizing as independent variables the climatological measurements of solar radiation, wind speed, outdoor relative humidity, and environmental temperature, and the working hours and occupancy of the building. The training, statistical comparison, and modeling performance were conducted through a computational methodology to obtain the best estimation architectures for each technique. The statistical parameter applied in the study were the root mean square error (RMSE) and the coefficient of correlation (R). Results reported the MLP as the technique with the best estimation accuracy (R=93.08% and RMSE=1.0166 for training, and R=92.90% and RMSE=1.1494 for testing), with an architecture composed by 6-30-1 (input variables, hidden neurons, and output value). Additionally, a sensitivity analysis identified the MLP and RBF as the techniques that best represent the physical behavior of the phenomenon studied. According to the sensitivity analysis, the most influential variables were the environmental temperature and outdoor relative humidity, followed by solar irradiation, working hours, wind speed, and the number of occupants. The proposed methodology represents an alternative method to simplify the analysis of thermal modeling in buildings exposed to tropical climates based on an experimental measurement approach. It can be applied for the development of smart sensors aimed at the efficient management of energy and thermal comfort in buildings. | |
