Reentry Vehicle Intelligent Trajectory Convex Optimization Method Based on Terminal Time Prediction

To address the efficiency problem of traditional sequential convex optimization (SCO) methods with uncertain terminal time for long-range hypersonic vehicle reentry, this paper proposes an improved method with neural network-based terminal time prediction strategy, which has distinctly higher computational efficiency than the traditional one. In the improved method, a neural network is used to fit the mapping between the vehicle’s current state and the terminal time, thereby replacing the parametric computation in the optimization process and thus improving efficiency. For network training, a large number of sample trajectories are first generated using the traditional sequential convex optimization method. Then, a multi-layer feedforward neural network is employed to approximate the mapping from the reentry vehicle’s flight states to the terminal time, thus completing the offline training. The simulation results demonstrate that the proposed algorithm reduces computation time by more than 50% compared to the SCO algorithm, satisfies the requirements for online trajectory generation, and can also adapt to special cases where the initial and terminal positions vary.