Developing an empirical predictive energy-rating model for windows by using Artificial Neural Network

In line with the growing global trend towards energy efficiency in buildings, this paper introduces a predictive model based on an artificial neural network model to rate the performance of windows in terms of saved heating and cooling loads. A four-story building representing the conventional type of residential apartments in Iran for four climates of cold, hot and humid, hot and arid, and temperate was selected for simulation. An artificial neural network model was developed based on ten variables of U-factor, SHGC, emissivity, monthly average dry bulb temperature, monthly average percent humidity, monthly average wind speed, monthly average direct solar radiation, monthly average diffuse solar radiation, orientation, and month as the input variables. The developed ANN model computes the amount of saved heating or cooling loads as a result of using a window with defined parameters. The best architecture of 10–10-1 with MAPE, RMSE, and R2 values of 1.4%, 0.008, and 0.985, respectively showed an acceptable predictive performance of the model. The predictions of this model, in line with the four levels of window performance defined in this paper, which range from excellent performance to weak performance, constitute the final rating of a window. The rated performance of the windows used in this study showed that the performance of a window can vary in cold and hot months, and windows should be rated according to the climate in which they are being used.