Data-driven model for cross ventilation potential in high-density cities based on coupled CFD simulation and machine learning

• Published in: Building and environment Vol. 165; no. C; p. 106394
• Main Authors: Ding, Chao, Lam, Khee Poh
• Format: Journal Article
• Language: English
• Published: Oxford Elsevier Ltd 01.11.2019; Elsevier BV; Elsevier
• Subjects: Aerodynamics; Building codes; Building design; Built environment; Computational fluid dynamics; Computer applications; Computer simulation; Coupled CFD simulation; Data-driven model; Density; Design; Early design support; ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATION; Fluid dynamics; Fluid flow; Ground level; High-density city; Hydrodynamics; Indoor environments; Learning algorithms; Machine learning; Model accuracy; Regression models; Simulation; Urban environments; Urban planning; Urban ventilation; Ventilation; Wind speed; Wind tunnel testing; Wind tunnels; Coupled CFD simulation; Early design support; Data-driven model; High-density city; Machine learning; Urban ventilation
• ISSN: 0360-1323; 1873-684X; 1873-684X
• DOI: 10.1016/j.buildenv.2019.106394

Effective urban ventilation through decent urban planning and building design can alleviate the deterioration of the urban built environment. However, natural ventilation requirements and guidelines in current building codes and standards are either qualitative or quantitative but subject to an absolute indoor airspeed threshold without considering the outdoor wind environment. To fill this gap, this paper develops an urban-scale coupled indoor and outdoor computational fluid dynamics (CFD) model and defines a novel ventilation index to assess natural ventilation potential. The index considers wind environments of both indoor and outdoor spaces. First, the coupled CFD model is developed to study wind-driven cross ventilation in high-density cities. A 3D isothermal CFD simulation is solved using the RNG k-ε turbulent model. The simulation results are compared with wind tunnel experiment data from literature. Second, six key design variables are used to generate 3840 parametric design variations for natural ventilation assessment. Third, a novel integrated index CIOIv (coupled indoor and outdoor interaction) is proposed to evaluate the wind speed ratio between the indoor area and outdoor reference area. For demonstration, CIOIv,F1 is used to represent CIOIv on the ground level. Data-driven CIOIv,F1 models are developed to predict indoor building ventilation potential for quick early design support. Compared with multivariate linear regression model, the Gradient Boosting non-linear model displays much higher prediction accuracy (mean absolute percentage error = 0.16 with R2  = 0.8). In early design stage, designers and engineers can skip the computational expensive CFD simulations and use this data-drive model to quickly check the building natural ventilation potentials of different design options in an urban environment. •An urban-scale coupled indoor and outdoor CFD model is developed to study cross ventilation.•Urban planning and building design features are parametrized and quantified using six key variables.•A novel ventilation evaluation index is proposed to integrate indoor and outdoor wind environments.•An advanced machine learning algorithm is used to develop data-driven CIOI prediction model for fast early design support.