Machine learning-enabled estimation of crosswind load effect on tall buildings

Abstract

This paper presents an approach to predict crosswind force spectra and associated response of tall buildings with rectangular cross-section based on machine learning (ML) technique and random vibration-based response analysis. An efficient ML algorithm, light gradient boosting machine (LGBM), was trained to predict crosswind force spectra of the tall buildings by using the database from the Wind Engineering Research Center at the Tamkang University embedded in the aerodynamic database of NatHaz Modelling Laboratory. Furthermore, an unsupervised ML algorithm, K-means clustering, was employed to advance the understanding of the crosswind force spectrum characteristics of the tall buildings. The effects of three factors, i.e., ground roughness, aspect ratio and side ratio, on the force spectra were discussed based on clustering. To predict the crosswind response of tall buildings, case studies were carried out to validate the predictive accuracy of the LGBM model combined with random vibration-based response analysis. The results demonstrate that the proposed method combined with the multiple database-enabled design module for high-rise buildings developed by the NatHaz Modelling Laboratory at the University of Notre Dame is effective and computationally efficient to provide fast and accurate predictions of the crosswind force spectrum and associated crosswind responses of rectangular tall buildings.