Research On UAV Path Planning Based On Reinforcement Learning In Mobile Edge Computing Syste

With the rapid development of modern communication and network technology,the number of devices located at the edge of network increases greatly.The great amount of data generated by these devices introduces a serious burden to the traditional cloud computing network.In this context,mobile edge computing(MEC)begins to show its advantages by providing services at the edge of network.However,the MEC servers in some remote areas are hard to deploy,so unmanned aerial vehicle(UAV)is considered in this article to build MEC servers and provide communication and computation services for edge users.To better satisfy the users’ service requests,this article adopts reinforcement learning algorithms to optimize UAV trajectories,so that the average task offloading amount of users and the system performance will be improved.Based on the above-mentioned work,this article carries out research and achieves following results:(1)For the UAV-mounted MEC system,the UAV path planning algorithms based on QLearning and Deep Q-Network are proposed due to the limited UAV battery capacity.The UAV trajectory is optimized under the energy consumption constraint.Moreover,the user fairness constraint is proposed to ensure the service quality for each user.Experimental results show that the proposed path planning algorithms can effectively improve the system performance.(2)In order to solve the unpractical problem that the users are stationary in some MEC systems,this section considers that users are moving in each time interval.Due to the dynamic change of the environment,the number of state-action pairs increases dramatically and traditional reinforcement learning methods cannot solve this problem any more.To alleviate this computation burden,a path planning algorithm based on Monte Carlo Tree Search is proposed.Moreover,a low-complexity scheme is further proposed to reduce the time cost in the iteration process.Experimental results show that the proposed path planning algorithms can achieve a larger average offloading tasks number than the benchmark algorithm.(3)Due to the limitations that single UAV has low service efficiency in multi-user scenarios,the MEC system with multiple UAVs is proposed in this section.A multi-UAV path planning algorithm based on Multi-Agent Deep Deterministic Policy Gradient algorithm is proposed to improve the number of average offloading tasks.Considering that UAVs may fly out of the service area or collide with other UAVs,a penalty function is associated with the reward function to better guide the flight of UAVs.Experimental results verify the effectiveness of the proposed scheme.Finally,all the research work of this paper is summarized,and the future work is discussed.