Airfoil Aerodynamic Coefficient Prediction based on Ensemble Learning

Abstract

The calculation of wing aerodynamic coefficient is the main content of airfoil design and research, which is of great significance to improve flight performance. The traditional methods to obtain the aerodynamic coefficients of the airfoil by computational fluid dynamics method or wind tunnel test have the disadvantages of large calculation and high test cost. In recent years, the high-speed development of machine learning has proved that it has strong nonlinear mapping ability. Therefore, more and more scholars apply it to the prediction of wing aerodynamic coefficients. The ensemble learning algorithm in machine learning has a strong ability of classification, regression and generalization ability. Taking this into account, Random Forest (RF) and Extreme Gradient Boosting (Xgboost), which are cutting-edge in ensemble learning, are applied to the prediction of wing aerodynamic coefficients for the first time. Xgboost has higher promotion potential than RF, so this paper additionally adjusts the parameters of Xgboost and obtains the optimal training parameters. Finally, we compare the prediction accuracy between non-ensemble and ensemble learning algorithms. The experimental results show that the ensemble learning algorithms have higher prediction accuracy than the classical regression algorithms. Among them, the best algorithm is Xgboost, and the prediction accuracy of RF is slightly lower than
Xgboost. The MAE, MSE, and RMSE of RF and Xgboost are approximately 10 ~ 100 times lower than that of other algorithms. In addition, Xgboost has lower time complexity and higher generalization capability.