Wind Pressure Prediction Of High Rise Buildings Based On Machine Learning

High-rise buildings are both tall and flexible,so they are more sensitive to wind loading.It is often necessary to conduct wind tunnel tests on the scaled high-rise building model to obtain the surface wind pressure coefficient for wind resistant design of tall building.Therefore it sometimes need to arrange a large number of wind pressure taps on the surface of the model to obtain the corresponding wind pressure data.Due to the limited available wind pressure scaning equipments in the wind tunnel test,it is impossible to arrange so many wind pressure taps and measure their wind pressure synchronizely.Therefore,it is particularly important to accurately predict the wind pressure for the points where no taps were arranged.In this paper,a variety of machine learning algorithms are adopted to predict the mean wind pressure coefficient,the RMS value for fluctuating wind pressure and power spectrum density of fluctuating wind pressure for the surface wind pressure of the isolated tall buildings.The main research contents of this paper are as follows:Firstly,the aerodynamic wind loading database for the isolated rectangular tall building conducted by Tokyo Polytechnic University,was adopted in this thesis for the basis data for the numerical prediction research on wind pressure on tall building.The time series of wind pressure for each wind pressure tap are obtained to calculate their mean wind pressure coefficient.Taking the coordinates of the wind pressure tap（X,Y）,the face number where the predicting tap was located,and the wind direction angle as the input paramters of the model,the mean wind pressure coefficient as the output of the model,four numerical predicting models based on the machine learning algorithm（including Decision Tree,Random Forest,Support Vector Regression and Deep Neural Network）were proposed in the paper respectively.The numerical prediction results for the the mean wind pressure coefficient,the prediction results for each pressure tap at the different wind directions,and the prediction results of different pressure taps in the same wind direction were analyze and compared.Secondly,the RMS value of fluctuating wind pressure coefficient of each measuring tap was obtained for the same aerodynamic wind loading database.Taking the coordinates of the wind pressure tap（X,Y）,the face number where the predicting tap was located,and the wind direction angle as the input paramters of the model,and the RMS value of the fluctuating wind pressure as the output of the model,four numerical predicting models based on the machine learning algorithm were also established respectively.The RMS value for each measurement pressure tap at the different wind directions,and the predicting RMS values for the different pressure taps in the same wind direction were analyze and compared.Lastly,by uiltizing the time history data of each windpressure tap in the aerodynamic wind loading database,the power spectral density of the fluctuating wind pressure for each wind pressure tap was constructed.The coordinates of the wind pressure tap（X,Y）,the face number where the predicting tap was located,the wind direction angle,and the frequency are selected as the input of the model,and the predicting power spectrum of fluctuating wind pressure was used as the output of the model,a deep neural network model was proposed in this thesis.The prediction results from of the deep neural network model in the wind direction of 0°,15°,30°and 45° were investigated ans compared respectively.The numerical prediction results show that the four machine learning algorithms can roughly predict the mean wind pressure coefficient and the RMS value of fluctuating wind pressure of the wind pressure taps in the high-rise building model,but the accuracies of prediction results are different for different algorithms and different measurement taps.The prediction results obtained from the deep neural network model are the most stable and accurate,then comes after the random forest algorithm,the accuracies of the numerical prediction results by the decision tree and the support vector regression model are more volatile.In general,the prediction results for the intermediate taps are better than those for the edge points,the accuracy of prediction results for the corner taps were the worst.Meanwhile the deep neural network algorithm can predict power spectrum density of the fluctuating wind pressure more accurately than other three algorithms.The accuracies of the numerical prediction results of the intermediate and edge taps are obviously better than those values of the corner taps.