基于强化学习与神经网络的动态目标分配算法

Dynamic targets assignment with reinforcement learning and neural network

针对传统的目标分配算法未考虑作战过程的实时变化情况,只按分配时刻的作战态势对多目标进行分配,导致火力单元分配过多或过少的问题。本文提出一种基于强化学习与深度神经网络的动态目标分配算法,根据不同想定剧情中的敌我目标状态,采用强化学习方式完成多步动态推演,利用专家经验和评估算法对分配数据进行评判,根据最优回报确立确定分配方案,通过利用训练好的深度神经网络为态势中的敌方目标分配我方武器进行的仿真实验结果可看出,与传统算法相比,本文算法在显著提升拦截成功率同时节省了分配时间。

Aiming at the problem that the traditional target assignment algorithm does not consider the real-time change of the combat process,it only assigns multiple targets according to the combat situation at the time of assignment,resulting in too many or too few firepower units.In this paper,a dynamic targets assignment with reinforcement learning and neural network is proposed.According to different scenarios,the state of the enemy and our targets is determined,and reinforcement learning is used to complete multi-step dynamic deduction.Finally,the assignment data is evaluated according to expert experience and evaluation algorithm,and the assignment scheme is established according to the final return.Through the simulation results of the trained deep neural network to distribute our weapons to the enemy targets in the situation,compared with the traditional algorithm,it can be seen that our algorithm can significantly improve the success rate of interception and save the assignment time.