Multi-Agent Reinforcement Learning Based Energy Consumption Minimization With Virtualization in Maritime UAVs-Assisted MEC Network

The flexibility and efficiency of unmanned aerial vehicles (UAVs) have been recognized as a viable solution to address the deficiency of deployment on the maritime mobile edge computing (MEC) networks. However, the limited computing resources of edge servers cannot support parallel processing tasks without performance loss. In this paper, we use virtual machine (VM) multiplexing technology to consider the performance degradation of parallel processing tasks in the UAVs-assisted MEC network. To ensure efficient task offloading, we focus on the trajectory planning and resource management for UAVs. Our objective is to minimize overall energy consumption while ensuring that time delays for all tasks are within tolerable limits. We first transform the joint optimization problem into a Markov Decision Process (MDP), then propose a Multi-Agent deep reinforcement learning-based Resource Management and Trajectory Planning (MA-RMTP) algorithm to solve the problem. The simulation results show that our proposed approach can significantly reduce the overall energy consumption compared to other benchmarks.