Explainable machine learning-based prediction for aerodynamic interference of a low-rise building on a high-rise building

Abstract

Interference effects between buildings may significantly change the wind pressure distribution on building façades and cause severe safety problems. In this study, a two-stage machine learning-based method was employed to investigate the interference effects of a low-rise building on a high-rise building. Firstly, the wind pressure coefficients on the building façades under various testing conditions (varies in height ratios, distances between two buildings and rotation angles of the interfering building) were predicted using three machine learning (ML) models: decision tree (DT), random forest (RF), and extreme gradient boosting (XGBoost). Among these models, XGBoost showed the superior performance and was selected as the basis for the explainable machine learning (EML) stage. Secondly, SHAP (SHapley Additive exPlanations) method was adopted to interpret the prediction results and analyze the importance of individual features. The SHAP explanations specified the weights contributed by features affecting the pressure coefficients, both qualitatively and quantitively. Generally speaking, SHAP analysis revealed that the building façade emerged as the most influential factor among all inputs, followed by the location on building façades, height ratio, and relative distance between the interference building and the principal building, while the rotation angle appeared to have the least impact. This study demonstrated that the two-stage ML-based method was effective in both predicting and explaining the complex interference phenomenon, and therefore could be promoted for tackling other intricate problems in wind engineering.