Machine Learning Models for Predicting Indoor Air Temperature of Smart Building

The indoor air temperature is one of the key factors to improve the performance of energy efficiency of buildings and quality of life in a very smart IoT environment. Therefore, a periodic and accurate prediction of the minimum and maximum indoor air temperature allows taking necessary precautions to handle the variations' impact and tendencies. During this assessment, we developed minimum and maximum indoor air temperature prediction models using multiple statistical regression (MLR), multilayered perceptron (MLP), and random forest (RF, where Rf is achieved once with tree depth 10 (RFdepth10), and once with tree depth 50 (RFdepth50)). The study was conducted at a building located within the University of Applied Sciences, Stuttgart, in Germany. Sensors were accustomed to aggregate data, which were used because of the input variables for the prediction. The variables are outdoor air temperature, indoor air temperature, humidity, and heating temperature. Performance of the models was evaluated with the coefficient of determination R-2 and therefore the root means square error (RMSE). The simulation results showed that the prediction by the MLP algorithm, based on minimum indoor air temperature models and also maximum indoor air temperature models, provides better accuracy with the very best R-2 and lowest RMSE in the independent test dataset. This survey developed a straightforward and powerful MLP model to predict the minimum and therefore the maximum indoor air temperature, which may integrate into smart building management system technology in the future.