Researsh On The Prediction Of Wind Pressure Distribution Based On LSSVM

With the development of buildings tend to be higher and more complicated, wind load becomes more and more important, and real-time access to wind pressure distribution under the actual wind direction becomes more realistic. At this stage, we can not get real-time wind pressure distribution by wind tunnel test, CFD simulation or field measurement with sensors. Here we use least squares support vector machine, and combined with CFD simulation data to find laws of wind pressure distribution under different directions, then realizing prediction of actual pressure distribution. The main contents include the following three aspects.Finished the establishment of prediction model for wind pressure distribution. Here we converted the regression prediction to optimization of the penalty and kernel parameters, and selected PSO to carry out parameter optimization. Known data includes CFD simulation and field measurement with sensors, that will be divided into training sample, the test sample and the sample to be predicted. The measuring points’ location coordinates and wind angles are considered to be factors of wind pressure, in order to eliminate the influen of wind velocity, we use shape coefficient instead of wind pressure.Verified the applicability and versatility of the prediction model for wind pressure distribution. Here we found two different shape buildings, r ectangular and curved building models, and get wind prerssure distribution data under 36 conditions(per ten degrees as a condition, o0,o10,……,o350 etc.), make prediction combined witn field measurement data. From the absolute error, the relative error and shape coefficient map, we can conclude that prediction model is applicability for wind pressure distribution, and versatility for different shape buildings.Achieved real-time wind pressure distribution of Zhuhai Opera House based on prediction model. It is necessary to real-time monitor as the harsh wind environment around Zhuhai Opera House. Here we formulated a detailed plan for real-time monitor wind velocity, direction and pressure. And making a reasonable plan for sensor placement based on prediction model. Before prediction, we already have CFD simulation data and field measurement data, then we pre dict wind pressure distribution based on above data. From the absolute error, the relative error and shape coefficient map, we can conclude that prediction model has good effects for Zhuhai Opera House. In addition, we analysis and explanation the fault tolerance for measured pressure values and measured wind directions by introducing random error.