Prediction of wind pressures on tall buildings using wavelet neural network

Abstract

For wind-resistant design of tall buildings, it is routine to obtain complete surface wind pressure distribution based on experimental data recorded at limited locations. The main goal of this study is to examine the usability of a hybrid artificial neural network method, i.e., Wavelet Neural Network (WNN), for simulating and interpolating wind pressures on tall buildings. Wind pressure measurement tests were carried out on two scaled tall building models. The performance of three different prediction models, namely back-propagation neural network (BPNN), genetic algorithm-back-propagation neural network (GA-BP), and WNN were examined for comparison purposes. The results are overall promising, in which all the three models were capable to reasonably replicating wind pressure characteristics (i.e., time series, power spectra and wind pressure coefficient distribution) using experimental data at selected locations. In particular, WNN was shown to produce the most satisfactory prediction results. This evidences that WNN can be considered as a useful and reliable tool to predict surface wind pressure on tall buildings. The outcomes of this study are expected to provide important implications to practically aid the wind-resistant design of tall buildings.