Antinoise Aerodynamic Parameter Estimation Approach for Hypersonic Vehicle Using Dynamic Equation and Flight Data

The flight data of hypersonic vehicle contain real aerodynamic characteristics that cannot be obtained from ground numerical simulation tests. Offline aerodynamic knowledge extraction based on flight data is meaningful but challenging. This paper presents an antinoise aerodynamic parameter estimation approach using dynamic equation and flight data to reduce the influence of random error and improve the estimation accuracy of the hypersonic vehicle offline aerodynamic model. For this purpose, the offline aerodynamic neural network (NN) model of a hypersonic vehicle is established. To enhance the antinoise ability of aerodynamic parameter estimation, the method takes the mean-square error between the actual pitch angle rate in flight data and the pitch angle rate based on the dynamic equation as the loss function, replacing the process of calculating the label value by the aerodynamic coefficient observation model. Furthermore, a Butterworth low-pass filter is introduced for flight state input noise processing. The significant advantage of this method is that the influence of noise error amplification on network correction is improved by avoiding the differentiation of flight data to time. Subsequently, the antinoise capability of the aerodynamic parameter estimation method is proved theoretically from two aspects of flight data input and parameter estimation output. Finally, for the hypersonic vehicle strong nonlinear aerodynamic model, the simulation results demonstrate that, in comparison with the current supervised learning approach, the aerodynamic parameter estimation method proposed in this paper can effectively improve the accuracy of the aerodynamic coefficient.