Prediction of Terminal Velocity in Reentry Phase for Hypersonic Aircraft Based on GRU Neural Network

Given the constraints of no-fly zones on predicting the terminal velocity of hypersonic boost-glide vehicles, this paper proposes a method for predicting the terminal velocity of the aircraft based on a gated recurrent unit (GRU) neural network. Firstly, an offline trajectory optimization is conducted by using a sequential convex optimization approach. A three-degree-of-freedom model for boost-glide vehicle reentry phase is established, and the original problem is transformed into a second-order cone programming (SOCP) problem through linearization, relaxation, and discretization operations. The SOCP problem is then solved using the convex optimization solver to obtain a series of optimal reentry avoidance trajectory data samples that can avoid randomly appearing threat areas. Secondly, with the state and control sequence of the trajectory as inputs and the velocity at the final moment as output, a neural network model is trained based on the optimal trajectory data sample library to obtain the best model for predicting the terminal velocity. Finally, the effectiveness of the velocity prediction algorithm is verified through ballistic simulation. The simulation results demonstrate that, under the condition of random threat zone positions, the designed GRU neural network model can achieve accurate prediction of the aircraft’s terminal velocity. It exhibits a satisfactory generalization capability to meet the accuracy requirements and has high computational efficiency.