基于强化学习的无人机智能任务分配方法

Reinforcement Learning-based Intelligent Task Assignment Method for Unmanned Aerial Vehicles

针对无人机群目标打击任务分配问题,提出一种基于强化学习的无人机智能任务分配方法。该方法提出一种任务分层框架,将多个无人机视为一个联盟并对目标进行分类,形成任务簇,并映射到无人机联盟中,通过多智能体强化学习算法(MADDPG)将任务簇内的目标与无人机联盟内的小无人机进行合理配对并对目标实施打击,得到MADDPG算法的回报值和飞行轨迹,并与DDPG算法、DQN算法的回报值和飞行轨迹进行对比。仿真结果表明,在小样本任务分配中,与不分层方法相比,该方法可以提高目标任务打击完成度,提升目标打击的效率;在分层框架下,相比于其他两种算法,收敛速度更快,收敛过程更加稳定。

Aiming at the task assignment problem of UAV swarm target strike, this paper proposes an intelligent UAV task assignment method based on reinforcement learning. This strategy proposes a task layering framework, which treats multiple UAVs as an alliance and classifies the targets to form task clusters, maps each task cluster to the UAV alliance. Through multi-agent reinforcement learning algorithm(MADDPG), the targets in the task cluster are reasonably paired with the small UAVs in the UAV alliance, then the targets are hit. The return value and flight path of MADDPG algorithm are obtained, and compared with the return value and flight path of DDPG algorithm and DQN algorithm. The experimental results show that in the task assignment of small samples, compared with the non-hierarchical method, this method can improve the completion degree of target task strike and improve the efficiency of target strike;under the hierarchical framework, compared with the other two algorithms, the convergence speed is faster, the convergence process is more stable.