Wind Interference Effects And Machine Learning Prediction Method Of Peak Wind Loads Of A Small Group Of Low-Rise Buildings

Due to the low height of low-rise buildings,the wind loads are easily affected by surrounding buildings.Compared with a large group of low-rise buildings,the wind loads of a small group of low-rise buildings might experience quite adverse effects,which might be greater than those of a large group of low-rise buildings and the isolated low-rise building.Hence,if the design wind loads of a small group of low-rise buildings are reduced according to the aerodynamic characteristics of a large group of low-rise buildings,it may result in an underestimation of wind loads.In the past,fewer studies focused on the interference mechanism of a small group of low-rise buildings.In addition,there are many factors affecting the wind loads of a group of buildings.If wind tunnel tests or numerical simulations are used to determine the wind loads acting on each group of buildings,it will consume much time and cost.Therefore,how to predict the wind loads of the target group of buildings based on the limited wind tunnel test samples is also an urgent issue to be investigated.The major contents and innovative achievements of this thesis are summarized as follows:(1)Wind tunnel tests were conducted to investigate interference effects among three low-rise buildings with flat roof.The influence of the building spacing on wind loads of buildings at different wind directions was discussed,and the results were compared with those of the small group of high-rise buildings and the large group of low-rise buildings in previous studies.When the wind direction is oblique,the shielding effects of the upstream building are significant on wind loads of the downstream building at the small building spacing.When the three buildings are arranged in tandem along the wind direction,the drag coefficient of downstream buildings increases and decreases rapidly at spacing ratios 0.4<S/W≤0.8,and has a large local peak at S/W=0.7.At S/W=0.7,the tap pressure time histories of downstream buildings are non-stationary with notable abrupt changes,but wind force time histories of downstream buildings are stationary.Then the large eddy simulations(LES)were conducted to simulate the flow field around the buildings.The results show that the above phenomena are attributed to a special time-averaged asymmetric wake regime between buildings.The asymmetric wake regime has two types of wake modes with symmetric bias flow directions,and two modes irregularly switch between buildings and the duration of each mode is ruleless.The above time-averaged asymmetric wake regime was not observed in previous studies on the small group of high-rise buildings and the large group of low-rise buildings.(2)LES was used to analyze the influence of the length-width ratio and height-width ratio of three tandem buildings on the asymmetric wake regime and the local peak of drag coefficients.The parametric analysis results showed that the influence of length-width ratio and height ratio of buildings was remarkable.When the height-width ratio is small,and the length-width ratio is less than 1.0,the asymmetric wake regime between the windward building and the central building becomes symmetrical,and there is no local peak in the variation curve of drag coefficients of the central building with the building spacing.In addition,the asymmetric wake regime is more sensitive to the change in the height-width ratio of buildings.When the height-width ratio is close to 0.8,the asymmetric wake regime among buildings disappears.(3)When a small amount of building group wind tunnel data is used to predict the peak wind loads of the target building group,it is difficult to construct an effective prediction model of peak wind loads of buildings directly due to the limited training data.Therefore,this study proposed a machine learning prediction method for predicting peak wind loads of a group of buildings combined with semi-supervised regression(SSR),extreme learning machine(ELM),and computational fluid dynamics(CFD)(named as SEC method).SEC method improves the prediction accuracy of peak wind loads of buildings in two ways.Firstly,the SSR method is used to expand the number of existing limited training samples.Secondly,the mean wind loads of buildings simulated by CFD are added as the input variable of the prediction model to enhance the linear mapping relationship between the input variable and the output variable in the prediction model.When the training of the SEC prediction model is completed,it can be directly applied to predict peak wind loads of the target building group of the same type as training samples without repeated training.Lastly,the SEC method was verified by the wind tunnel tests data of three low-rise buildings.The results show that when the wind loads data of building groups is limited,the prediction accuracy of the SEC method is higher than that of the prediction model trained directly based on the limited training data.In addition,when the predicted target building group is in the local range where the peak wind loads alter rapidly with the variation of influence factors(such as building spacing and wind direction),the advantages of the SEC method are more obvious.(4)Based on the multiple sampling results of wind tunnel tests,minimum pressure coefficients on the roof of buildings were analyzed by using the extreme probability distribution method and unbiased estimation algorithm.The scheme of the zone division on the roof was determined according to the distribution law of the most critical minimum wind pressure coefficients on the roof of buildings.Then,for each zone,the variation of the most critical minimum wind pressure coefficients among all wind directions with building spacing was investigated.Considering the incomplete correlation of fluctuating wind loads among different pressure taps on the roof,the formulas to calculate the design suggested values of area-averaged most critical wind pressure coefficients of all roof zones were proposed as a function of the tributary area.These formulas can be used as the reference for determining wind loads during the design of roof claddings of such low-rise buildings at different building spacings.