Prediction of Wind Pressure Coefficients in Building Energy Simulation using Machine Learning

Abstract

A common source of uncertainty for the airflow rate calculation in building energy simulation software is the use of surface-averaged generic wind pressure coefficients instead of local wind pressure coefficients for any building shape. This work explores the calculation of local wind pressure coefficients using artificial neural networks for a specific building shape. Results show that it is possible to approximate the local wind pressure coefficient using artificial neural networks with good agreement with the experimental data (up to 50% reduction in the error when compared to Air Infiltration and Ventilation Centre (AIVC) database Cp data). Subsequently, local wind pressure coefficients obtained from the neural networks are introduced to corresponded building energy simulation software. Simulations were performed for a case study in order to compare the performance obtained using data from (a) wind tunnel experiments (reference best practice), (b) surfaceaveraged generic wind pressure coefficient from AIVC and (c) artificial neural networks local wind pressure coefficient. Artificial neural network simulated air flow rates show significant improvement in accuracy when compared to AIVC.