Identification of Aerodynamic Parameters Using Improved Physics-Informed Neural Network Framework

An on-line aerodynamic parameters identification method is proposed based on improved Physics-Informed Neural Network (PINN) to address aerodynamic parameters error problem during flight control. An integration-based loss function is utilized to ensure that the neural network can learn the correct physical equation information, and adopts a parallel neural network architecture to reduce network complexity. To ensure the feasibility of the network, the input and output data are measurable by the Integrated Navigation System. The improved PINNs is used to identify the aerodynamic parameters of the Reentry Gliding Vehicle in numerical simulation. Simulation results demonstrate that the network can effectively identify aerodynamic parameters during the flight process and the proposed method is insensitive to measurement noise. The proposed method can provide information for the design of multi constraints guidance laws for flight vehicle.