关键词: 拦截弹, 攻击区, 环境干扰, 径向基神经网络, 自注意力机制

Abstract: An RBF+SA attack zone network model that uses radial basis function（RBF）neural network as the base model and integrated with self-attention（SA）mechanism, is proposed to address the online calculation problem of the launch envelope when the interceptor missile intercepting a hypersonic gliding target under complex environmental disturbances. Initially, motion equations for the interceptor missile considering environmental disturbances such as wind and atmospheric density perturbations are established, and a launch envelope based on the given target motion, interceptor missile maneuverability, and guidance strategy is defined, characterized by the initial positions of the interceptor missile and the initial velocity direction of the interceptor missile. Additionally, sample data for the attack zone are obtained offline. Then considering the strong nonlinearity of the impact of multiple environmental disturbances on the interceptor missile’s attack zone, base models for the attack zone under single disturbances (e.g., wind and atmospheric density) using RBF neural networks are established, and a dynamic integration model of SA mechanism network is developed to dynamically integrate single-disturbance attack zone base models across various combat scenarios to achieve fast training speed and high accuracy. The results demonstrate that the RBF+SA model accurately represents the attack zone under multiple complex environmental disturbances with high precision and achieves higher prediction accuracy using fewer samples.

Key words: interceptor missile, attack zone, environmental disturbances, radial basis function neural network, self-attention mechanism