Study On Trajectory Planning And Resource Allocation Of Maritime Multi-UAV-assisted Mobile Edge Computing System

With the rapid development of maritime activities,various computation-intensive applications are showing a booming trend,which puts increasing demands on maritime communication network.The current maritime communication network cannot meet the wireless communication requirements of big data,high-speed and cost-effective.Intelligent integration of traditional maritime communication network is an inevitable choice for maritime network construction.With this as the original intention,the research on maritime communication network is carried out in this paper.This paper introduces mobile edge computing(MEC)technology to expand the existing maritime communication architecture,in order to provide real-time and efficient computing services for maritime terminals of resource scarcity or latency sensitive.Due to the unique geographic consideration,climatic conditions and user distribution of maritime environment,it is extremely difficult and expensive to deploy MEC architectures.Unmanned aerial vehicles(UAVs)have the advantages of flexibility,mobility,cost-effective and easy deployment,so adopting a UAV-assisted MEC architecture to provide powerful computing capabilities for maritime terminals is an effective solution to this problem.Considering the limited computing resources and battery capacity of UAVs,it is risky to complete computing services independently.It is of great significance to study the collaborative offloading and resource sharing among UAVs in the multi-UAV-assisted MEC system.The main contributions of this paper are as follows:In the multi-UAV-assisted MEC maritime scenario,a two-layer UAV-assisted MEC architecture is established,which consists of a centralized top-UAV(T-UAV)and a group of distributed bottom-UAVs(B-UAVs).This collaborative UAV-assisted MEC network architecture,as a supplement for central cloud,not only provides efficient computing power and reduces task completion time,but also avoids the latency and energy consumption of long-distance transmission from maritime terminals to central cloud,at the same time can reduce network congestion.A basic technology to implement MEC is virtual machine(VM)multiplexing,which allocates multiple VMs in a same physical machine to realize multi-task parallel computing.Compared with the existing literatures,this paper considers the problem of parallel computing in a more practical MEC scenario.It solves the problems of I/O interference among VMs and the parallel computing of tasks with the same amount of data in different VMs.In the MEC system with I/O interference,we optimize the number of VMs according to the amount of data for each computing task to minimize the parallel computing latency.This paper study the joint optimization problem of T-UAV trajectory optimization and VMs resource allocation under the TDMA and FDMA methods respectively.T-UAV and B-UAVs have different access methods for wireless transmissions,and correspondingly establish different mathematical models and reward functions,which will produce different training effects.Considering that this optimization problem has many practical constraints and non-convex characteristics,it is difficult to solve it by traditional methods.We model it as a Markov decision process(MDP),and solve it with the deep Q-network(DQN)algorithm and the deep deterministic policy gradient(DDPG)algorithm to solve the optimal strategy of this optimization problem and achieve the goal of minimizing the system average latency.The simulation results show that compared with other benchmark algorithms,the DQN algorithm and DDPG algorithm proposed in this paper can significantly reduce the system average latency regardless of the access method,but the DDPG algorithm is more effective in improving the system latency performance.Specifically,the latency performance of two algorithms in the FDMA method is slightly better than that in the TDMA method,but the transmission energy consumption is much higher than that in the TDMA method.