Prediction Of Wind Loads On High-rise Buildings By Machine Learning Based Algorithms

In recent years,buildings are developing towards a higher and more flexible trend with the rapid development of society and technology and the impact of wind on buildings is also increasing.Wind tunnel test and computational fluid dynamics(CFD)are the two common methods usually used for studying the wind-induced response and wind load of high-rise buildings.Machine learning enable computer systems to find hidden laws from a large amount of data to make accurate predictions.Machine learning showed broad application prospects once it came out,and many researchers applied machine learning to a series of engineering problems.The machine learning algorithms have been adopted to predict the wind load of highrise buildings in this study,the main contents as following:(1)Machine learning algorithms have been adopted to predict wind pressures on standard tall building model.The input of the machine learning model includes turbulence intensity,scale ratio,incident wind angle and the position of measuring point,and the output is the mean wind pressure coefficients and the fluctuating wind pressure coefficients.Wind pressure data sets of standard tall building model are collected from a series of wind tunnel experimental studies and used to train four machine learning algorithms including ridge regression,decision tree,random forest and gradient boosting regression tree.The principle of the four algorithms are described in detail and the cross-validation is utilized to optimize hyperparameters of different algorithms.The gradient boosting regression tree algorithm has been proved to predict the mean wind pressure coefficients and the fluctuating wind pressure coefficients of standard tall building model well by comparing the prediction performance of the four machine learning algorithms on the test dataset.(2)Machine learning algorithms have been adopted to predict wind force coefficients on standard tall building model.The input of the machine learning model is turbulence intensity,wind directions and the heights of measurement layer,and the output is the mean drag force coefficients,RMS drag force coefficients and RMS lift force coefficients.A large amount of wind force coefficient cases are collected for training and testing the four machine learning algorithms including K-nearest neighbor,support vector machine,gradient boosting regression tree and XGBoost by conducting wind tunnel tests.The Tree-structured Parzen Estimator and cross-validation is utilized to optimize hyperparameters of the algorithm.By comparing the prediction performance of different machine learning algorithms in the test dataset,the results show that the gradient boosted regression tree can predict the mean drag coefficient and RMS lift coefficient of standard tall building model well,while XGBoost can predict the RMS drag coefficient of standard tall building model well.The prediction error is within 20%.(3)Machine learning algorithms have been adopted to predict power spectrum on standard tall building model.The input of the machine learning model is turbulence intensity,wind directions and the reduced frequencies,and the output is the power spectrum of the alongwind,across-wind,and torque base moment coefficients.The power spectrum data of the wind tunnel test are used to train four machine learning algorithms including gradient boosting regression tree,histogram gradient boosting regression tree,XGBoost and neural network.The hyperparameters of the algorithms were optimized by the Tree-structured Parzen Estimator and cross-validation.By comparing the prediction performance of the four algorithms in the test set,it is found that the gradient boosting regression tree can predict the power spectrum of the base moment of standard tall building model well and the correlation coefficient between the predicted value and the experimental value is not less than 0.97.The study shows the feasibility of machine learning to predict the wind pressure coefficient,wind force coefficient and power spectrum of standard tall building model,and provides a reference for applying machine learning to wind-resistant design of high-rise buildings.