Modelling enhancement of cross-ventilation in sheltered buildings using stochastic optimization

•A new method is proposed to find the optimum values for the closure coefficients of RANS models.•The method is based on the stochastic optimization and Monte Carlo sampling technique.•The method is applied to improve the CFD accuracy for cross-ventilation of a sheltered building.•The airflow rate error has decreased from 70% to 8% using the modified closure coefficients. Accurate representation of turbulence phenomenon in Computational Fluid Dynamics (CFD) modeling of cross-ventilation around and inside buildings is a challenging and complex problem, especially under the sheltering effect of surrounding buildings. Steady Reynolds Averaged Navier-Stokes (RANS) models are broadly used in many practical applications. However, these models mainly fail to predict accurate distribution of flow characteristics in the cavity formed between the buildings, and hence miscalculate the attributed cross flow and airflow rate through buildings. In this study, a novel and systematic methodology is proposed to enhance the accuracy of the k-ε model for the urban study applications such as cross-ventilation in the sheltered buildings. A microclimate CFD model for a case study of a cross-ventilation experimental work by Tominaga and Blocken (2015) was firstly constructed and validated. In the next step, the closure coefficients of the k-ε model were modified using a stochastic optimization and Monte Carlo Sampling techniques. The probability density function (PDF) of all closure coefficients were given to the optimizer and proper objective function defined in terms of different validation metrics. The modified coefficients obtained from the developed systematic method could successfully simulates the cross-ventilation phenomena inside the building with an airflow rate prediction error less than 8% compared to the experiment while other RANS models predicted the airflow rate with up to 70% error. The effectiveness of the optimization technique was also discussed in terms of validation metrics and pressure coefficients.