基于博弈论和强化学习的无人机电力巡检卸载策略

Power Inspection and Unloading Strategy of UAV Based on Game Theory and Reinforcement Learning

无人机组在执行应急故障巡检时由于受到设备本身计算能力和能源资源的限制,无法较好地执行密集型计算任务。针对这一问题,该文提出一种基于博弈论(game theory,GT),与强化学习的卸载策略,该策略在无人机之间建立非合作博弈将最小化成本函数定义为能量开销和延迟组合,同时证明了至少一个纳什均衡(Nashequilibrium,NE)的存在性,提出了一种分布式算法求解博弈双方的NE解,在此基础上,通过基于随机学习自动机(stochastic learning automata,SLA)理论的强化学习方法,实现无人机对边缘服务器的有效选择。仿真结果表明,与其他卸载策略相比,该文所提的卸载机制对降低无人机能耗、系统成本和网络时延效果显著。

Due to the limitation of the equipment computing capacity and the energy resources,UAVs cannot perform intensive computing tasks well.In order to solve this problem,this paper proposes an unloading strategy based on the game theory and the deep learning.This strategy establishes the cooperative games between the unmanned aerial vehicles(UAV),so the minimized cost function is defined as the energy costs and delays,and the existence of at least one Nash equilibrium(NE)is proved.A distributed algorithm for solving the NE of both the game sides is proposed.On this basis,through the reinforcement learning method based on the theory of SLA,the UAVs effectively realizes the implementation of the edge server option.Simulation results show that compared with the other unloading strategies,the unloading mechanism proposed in this paper has a significant effect on reducing the UAV energy consumption,the system costs and the network delays.