Intention prediction of UAVs based on improved DDQN

Abstract

Intention prediction plays an indispensable role in future informationized air combat, which will help the command and control center to make a more correct decision. In this paper, an intelligent intention prediction approach based on an improved double deep Q network (DDQN) is developed to generate real-time intention flight paths for unmanned aerial vehicles (UAVs) in complex air combat environment. Initially, by introducing an actual topographic map, the different threats of UAVs in different terrains on the map are analyzed, based on which, a terrain environment reward function is constructed. Secondly, by splitting a complete maneuver action into six basic maneuver units and determining the probability value of each unit, a maneuver reward function is obtained. Further, to improve the real-time performance and accuracy of the standard DDQN algorithm, an improved DDQN algorithm is proposed by using temporal-difference (TD) method and binary tree data structure. Finally, the simulation results verify show that the proposed algorithm has achieved better results under complex terrain conditions.