Aerodynamic Fusion Prediction of Similar Airfoils Based on Conditional Distribution Alignment

Aerodynamic performance prediction is a crucial aspect of aircraft design, performance optimization, and safe operation, playing a vital role in the aerospace field. Due to the scarcity and high cost of wind tunnel test data, this paper proposes a transfer learning method combining the Ada-BoostTree aerodynamic prediction model and Latent variable represented Conditional Distribution Alignment. First, latent variables are used to represent the distribution differences of wind tunnel data for similar airfoil shapes. Then, the latent variables are optimized by matching the conditional distributions across domains, thereby aligning the global distribution differences. Finally, aerodynamic data from different airfoils are aligned and subsequently used to train the AdaBoost Tree model. Experimental results demonstrate that this method can effectively improve aerodynamic prediction performance when wind tunnel data samples are limited.