Research On Formation Method And Formation Transformation Of Multi-UAV In Adversarial Environment

When the function of a single UAV(Unmanned Aerial Vehicle)is still limited,in the face of complex combat missions and unpredictable combat environments,the performance possessed by a single UAV has gradually failed to meet the operational requirements.UAVs perform more comprehensive missions in the form of multi-aircraft cooperative operations.Multi-UAV formation is an important part of multi-UAV system,which is the premise of task allocation and path planning.It is confronted with great challenges in the current highly adversarial dynamic environment:(1)Multi-UAV formation constructed by existing formation methods cannot be satisfied simultaneously in terms of formation stability and formation transformation autonomy.(2)When the formation is affected and needs to be adjusted,the formation transformation is slow,and the flight path is overlapping and overlong.In order to solve these problems in the direction of formation of multi-UAV,this paper has carried out some research as follows.Firstly,in this paper,an improved leader-follower method is proposed by studying the classic multi-UAV formation method and combining some characteristics of leader-follower method and virtual structure method.In this method,the improved artificial potential field algorithm is used to provide basic formation control for multi-UAV formation.The idea of distributed consistency algorithm is introduced to construct formation,which provides guarantee for information consistency of multi-UAV system.The dependence of formation on ground control center is reduced and the autonomy of the UAV formation is improved by keeping the leader aircraft as the decision center.The safety of UAV formation is improved by preparing multiple physical leader aircrafts to form a multi-core formation mode.This method solves the problem that the safety and the autonomy of formation transformation of multi-UAV cannot be simultaneously met in adversarial environment.Secondly,based on multi-agent reinforcement learning algorithm and heuristic function,this paper proposes an improved heuristically-accelerated reinforcement learning algorithm to solve the problem of multi-UAV formation transformation.The algorithm improves the convergence efficiency of reinforcement learning by a heuristic back propagation algorithm suitable for formation transformation.Combined with simulated annealing related methods,the action selection strategy in reinforcement learning is improved so that reinforcement learning algorithm can obtain high-quality solutions at a higher speed.This algorithm solves the problem of low formation transformation efficiency of multi-UAV in adversarial environment.Finally,this paper uses technologies of virtualization and distributed system to realize the simulation experiment of multi-UAV formation.The simulation results show that the improved formation method of leader-follower meets the requirements of combat missions and has advantages in stability and autonomy.The improved heuristically-accelerated reinforcement learning algorithm proposed in this paper solves the problem of formation transformation effectively and provides a safe and high-quality flight path for the formation transformation of multi-UAV.