The Research And Application Of Aerodynamic Data Analysis Methods

With the rapid development of aerodynamics,a variety of heterogeneous aerodynamic experimental data continue emerging.Faced with these data,how to conduct effective data analysis and mining has become a puzzle for aerodynamics experts.Based on this problem,this paper analyzes the data mining of aerodynamic data deeply and designs the corresponding prototype system of aerodynamic data platform.Through analyzing the existing aerodynamic data,the data mining of representative structured data and semi-structured data are carried out.By analyzing the existing aerodynamic data platform,a prototype design of aerodynamic data platform is put forward which store aerodynamic data effectively in the background of big data.Specifically,the following works are completed:1,Select the representative semi-structured data(finite element surface data)for data mining analysis through the proposed extreme random forest for predicting aerodynamic surface flow field files.Experiments show that the overall performance of the proposed method is superior to the existing methods;2,The neural network trained by quasi-Newton method is used to predict the aerodynamic coefficient.Experiments show that it reduces the model training time greatly while maintaining the predictive ability compared with the previous studies-the network trained by gradient descent method;the over-fitting phenomenon of the prediction process is suppressed,making the generalization ability of the model prediction of aerodynamic coefficient to enhance;3,The research work above is integrated and a design of aerodynamic data platform is completed.A storage design of effective stored aerodynamic data in the background of big data is proposed.