The strong and unsteady wind imposes severe challenges to the safe flight and aerodynamic prediction of the fixed-wing aircraft. Traditional aerodynamic models established in the wind-oriented coordinate system have a clear physical meaning but cannot be readily applied to unsteady windy environments. This paper proposes an innovative "neural"-fly aerodynamic modeling method based on deep meta-learning to accurately predict the aerodynamic forces and moments online for fixed-wing aircraft subjected to strong and unsteady wind. Based on variables in a coordinate system relative to the ground, this method decomposes the aerodynamic forces and moments into the sum of polynomial multiplication and constructs the common aerodynamic base functions by a three-step deep meta-learning algorithm using the Generative Adversarial Network. The application of the method for the fixed-wing aircraft F-18 demonstrates that the method can accurately predict the aerodynamic forces and moments under unknown wind conditions, laying a good foundation for real-time aerodynamic modeling.