Parameter Identification of Hypersonic Vehicle Based on LSTM Algorithm

Abstract

Hypersonic vehicle(HSV) has the advantages of wide range of flight height and strong long-range flight capability, but large range of aerodynamic pressure changes cause great changes in aerodynamic parameters of aircraft model, which directly affects the navigation and control of the system. In this paper, the deviation value of aerodynamic parameters in flight is identified based on the Long Short-Term Memory(LSTM) algorithm, which can be directly used in controller. Meanwhile, the real-time calculation of aerodynamic parameters can be realized by combining with the undeviated aerodynamic parameters preloaded by aircraft. Compared with the cyclic neural network, the algorithm increases the forgetting coefficient and avoids the gradient explosion and gradient vanishing problems. At the same time, a fully connected neural network is added behind the LSTM network to make the identification results more accurate.Firstly, the kinematics model of aircraft is established to determine the required identification parameters, and training data is generated based on the model. Then, the parameter identification network connected with LSTM network and single hidden layer fully connected network is constructed, and the data is input for network training. Finally, the trained network is used for parameter identification. The simulation results show that the identification error is less than 10%, which is of great engineering application value.