LAUSR

The indoor environment has a significant impact on our wellbeing. Accurate prediction of the indoor air distribution can help to create a good indoor environment. Reynolds-averaged Navier-Stokes (RANS) models are commonly used for indoor airflow prediction. However, the Boussinesq hypothesis used in the RANS model fails to account for indoor anisotropic flows. To solve this problem, this study developed a data-driven RANS model by using a nonlinear model from the literature. An artificial neural network (ANN) was used to determine the coefficients of high-order terms. Three typical indoor airflows were selected as the training set to develop the model. Four other cases were used as testing sets to verify the generalizability of the model. The results show that the data-driven model can better predict the distributions of air velocity, temperature, and turbulent kinetic energy for the indoor anisotropic flows than the original RANS model. This is because the nonlinear terms are accurately simulated by the ANN. This investigation concluded that the data-driven model can correctly predict indoor anisotropic flows and has reasonably good generalizability.