Deep Learning Midcourse Guidance for Interceptor Missile

Abstract

A midcourse guidance method of interceptor missile based on Long Short-Term Memory deep learning networks is studied in this paper. Comparing with the guidance method using traditional neural networks, the miss distance of this method is significantly reduced. In the simulation process, the real-time states of interceptor missile are taken as the inputs of deep learning networks, and the trajectory integration is carried out with the output vector. Moreover, the guidance method is improved by changing three characters: the density of the selected sample trajectory, the size of the sample airspace and the size of the simulation airspace. Also, simulations of the trajectories pointing to the random prediction intercept points selected in a certain simulation space are carried out. Different deep learning guidance rules should be selected according to different application conditions.