基于自编码网络的导弹攻击区实时计算方法

针对传统中远距空空导弹三线攻击区无法为飞行员提供丰富的战术决策信息，火控系统计算攻击区实时性差、精度低的问题，提出以攻击机为中心，考虑目标逃逸机动的新型导弹杀伤包线概念。分析经典战例中目标机规避导弹的常见机动方式，将攻击区抽象为导弹的七种杀伤包线，给出准确的计算方法并进行离线仿真。确定8种影响杀伤包线的运动参数，构建样本库。引入深度学习方法，建立降噪自编码网络（AE）模型，采用无监督学习提取样本初级特征，获取表征样本库非线性规律的高维特征量；建立深度网络模型，采用监督学习提取高维特征量中的高级特征并进行拟合。实验表明深度网络的拟合值与六自由度仿真结果以及导弹真实数据相比，误差可控制在15 m之内；网络在线解算只需0.04 s，能够满足实时性需求；新型杀伤包线为飞行员及时掌握敌我态势提供了有效的辅助信息，为机动决策提供理论依据。

Real-time calculation of missile launch envelope based on auto-encoder network

Traditional three-line attack area of medium-range air-to-air missile can’t provide effective tactical decision information for the pilot, and it has poor time effectiveness and low precision in calculation using fire control system. To address these problems, a new type of missile launch envelope, under the consideration of the target escape maneuver, is proposed with the attack aircraft as the center. In this paper, common and effective tactical maneuvers, which are used to escape from missiles successfully, are considered and analyzed through many classic examples of battles. The attack area is abstracted into seven killing envelops. the calculated methods are given and the offline simulations are carried out. The sample database is established by identifying eight motion parameters and envelop-values. By introducing deep learning methods, the Auto-Encoder (AE) network model is constructed. By adopting unsupervised learning, the primary features of the sample are extracted, and the high dimensional vectors describing nonlinearity in the sample database are obtained. The deep network model is established with high dimensional vectors to extract advanced features for fitting. Experimental results show that the loss of the deep network can be controlled within 15 m compared with the six-degree-of-freedom simulation results and the actual missile data. The online calculation time is only 0.04 s, which can meet the real-time requirements. The new missile envelope can provide timely and effective information for pilot to understand enemy situation and support maneuvering decision-making.