The application of improved BP neural network in booster wreckage impact point prediction
Year: 2025
Type: conference-paper
Abstract: During flight test missions of rockets or missiles, the process of booster wreckage descent is influenced by a variety of complex factors. In response to the issues of long solution time and low accuracy of traditional wreckage impact point prediction methods, an improved BP neural network impact point prediction model is proposed. This model, based on the lens imaging reverse learning strategy and the Subtraction-Average-Based optimizer, has improved the issues of BP neural network being prone to local optima and poor adaptability. At the same time, by detailed modeling and simulation of factors such as wind fields, atmospheric models, and aerodynamic damping torque, a comprehensive and diverse wreckage impact point database has been constructed. This has enabled reasonable training of the neural network, which has been compared with various heuristic algorithms in simulation tests. Simulation results indicate that the algorithm can effectively predict the impact points of booster wreckage under different separation conditions, with strong stability, small error, and short time consumption, providing a reference for the safe implementation of missions.
Authors: Guodong Cai, Jiankai Qin, Yuelong He, Yudong Zhang
Institution: China Academy of Aerospace Aerodynamics
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W4408614055