Integrating neural network models with computational fluid dynamics (CFD) for site-specific wind condition

Most building energy simulations tend to neglect microclimates in building and system design, concentrating instead on building and system efficiency. Energy simulations utilize various outdoor variables from weather data, typically from the average weather record of the nearest weather station that is located in an open field, near airports and parks. The weather data may not accurately represent the physical microclimate of the site, and may therefore reduce the accuracy of simulation results. For this reason, this paper investigates utilizing computational fluid dynamics (CFD) with neural network (NN) model to predict site-specific wind parameters for energy simulation. The CFD simulation is used to find selected samples of site-specific wind conditions. Findings from CFD simulation are used as training data for NN. A trained NN predicts site-specific hourly wind conditions for a typical year. The outcome of the site-specific wind condition from the neural network is used as wind condition input for the energy simulation. The results of energy simulation using typical weather station data and site-specific weather data are compared in this paper, in order to find the possibility of using site-specific weather condition by NN with CFD to yield more realistic and robust ES results.