Machine Learning Modeling Strategies for Drag Assessment in Suction Flow Control Design
To develop a high-precision modeling strategy for drag assessment in suction flow control,five machine learn-ing models are selected:two Gaussian Process Regression(GPR)models based on Radial Basis Function(RBF)kernel and Matérn kernel,two Support Vector Regression(SVR)models based on RBF kernel and Sigmoid kernel,and one Ker-nel Ridge Regression(KRR)model based on RBF kernel.Using the NACA0012 airfoil as a case study,the modeling ef-fects of different hyperparameter evaluation metrics,modeling variables,sampling methods,and machine learning models are compared.The study on modeling variables shows that compared to direct modeling of the drag coefficient,modeling the drag reduction increment in suction flow control can reduce the mean absolute error of regression prediction by 24％to 62％.Using the drag coefficient of the airfoil under non-suction conditions as an input variable can reduce the mean ab-solute error of regression prediction by 43％to 86％.The overall results indicate that machine learning modeling for drag assessment in suction flow control design is feasible.
suction flow control technologyflow controlmachine learningaerodynamic characteristic modelingmodel-ing strategies
万方数据
维普
