Prediction of the indoor airflow temperature distribution with a heat source using a multilayer perceptron

Few studies have considered using artificial neural networks to predict the temperature distribution inside a room with a heat source, even though such an approach is potentially much faster than numerical analysis. In this study, a multilayer perceptron (MLP) model was developed and applied to predicting the temperature distribution in a room with a heat source, where the input variables were the x- and y-coordinates and the distances from the bottom of the wall to the airflow inlet and outlet. The MLP model was trained and evaluated to optimize the hyperparameters (i.e., batch size, node, learning rate, and number of layers) for the prediction accuracy. When compared with a computational fluid dynamics (CFD) simulation, the MLP model showed comparable prediction accuracy, but the computational time was about 11 s compared with 27 min for the CFD simulation. Thus, the MLP model can predict the temperature distribution in a room with an internal heat source with high accuracy and very low computational cost.