Refined prediction of crosswind forces on rectangular tall buildings with different side ratios

Abstract

The crosswind responses of tall buildings are complicated due to numerous influencing factors. However, the relevant codes for estimating the crosswind responses have a limited application scope and the corresponding empirical formulas may cause inaccurate assessment of structural responses due to the fitting errors. This study aims to achieve more accurate and flexible predictions of the crosswind forces and wind-induced responses of rectangular tall buildings in comparison with using the conventional empirical formulas. A series of high-frequency force balance (HFFB) tests was conducted on rectangular tall buildings with side ratios from 0.2 to 5 to investigate the crosswind force spectra and the root mean square (RMS) base moment coefficients in two terrain categories. The crosswind force spectra and the crosswind responses of the structure were predicted by using the new proposed formulas and genetic algorithm-backpropagation (GA-BP) neural network, respectively. The results showed that the RMS crosswind base moment coefficients of rectangular tall buildings generally increase with the increase in the side ratio, and can be expressed as piecewise functions related to the side ratio. The proposed formulas can effectively predict the crosswind force spectra of rectangular tall buildings with different side ratios. Compared with the traditional nonlinear fitting method, the GA-BP neural network has higher fitting accuracy and stronger ability for extrapolation prediction. Combined with the random vibration theory, the GA-BP neural network can accurately predict the crosswind responses of rectangular tall buildings. By comparing with the relevant codes, it shows that the crosswind responses of rectangular tall buildings under different side ratios are overestimated to different degrees by GB 50009-2012.