Research on Maneuvering Decision of UCAV with Deep Q-network

In the context of intelligent air combat, the use of UCAVs to complete military operations is a current research hotspot. The autonomous maneuvering decision capability of UCAVs determines the winning and losing outcome of aerial combat. In order to study the problem of maneuvering decision-making in 1V1 air combat of UCAVs, this paper presents a maneuvering strategy generation algorithm for UCAVs based on deep Q-network. The environment in autonomous air combat is complex and variable. This paper firstly establishes a three-dimensional air combat situation and UCAV maneuvering model to satisfy the simulation research. According to the air combat situation assessment scheme, the variable weight theory is introduced to design the dynamic adjustable reward reshaping function, and a network model is trained by deep Q-network to make maneuvering decisions, so as to complete the autonomous combat. The results of simulation experiments show that the UCAVs is able to perform the perception of the current airspace situation under the given initial conditions. The maneuvering actions given by the maneuvering decision algorithm can increase the dominance value of UCAVs and maintain the dominant state, improving the maneuvering decision capability.