Deep Neural Networks-Based Hypersonic Vehicle Reentry Impact Point Prediction With Fixed-Length Lightweight Training Data

Hypersonic vehicles play a key role in aerospace applications with their speed, maneuverability, and stealth. Accurate Impact point prediction is crucial for improving system performance. However, existing impact point prediction networks face challenges such as redundancy in input-output pairs and fixed input lengths, which cause weak feature correlations and data wastage. In response to these challenges, this paper proposes an intelligent impact point prediction (IPP) method based on dynamic data processing, focusing on key issues such as flight state selection optimization and time series feature extraction. First, a three-degree-of-freedom re-entry model for the aircraft is established, and a trajectory database is constructed as the basis for neural network training. Then, a network mapping method based on correlation analysis is proposed, selecting flight state variables highly correlated with the impact point prediction as the neural network input, thus reducing redundant information. Following this, a dynamic data processing method based on time decay factors and feature change rates is introduced, which can adaptively select important time-step data to enhance the effective use of temporal information. Finally, an impact point prediction network is constructed based on Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM). Simulation results show that the intelligent impact point prediction method proposed in this paper can accurately predict the impact point, with the impact point prediction error within a range of several kilometers.