Research On Computation Offloading And Resource Allocation In UAV-enabled Mobile Edge Computing System

Mobile Edge Computing(MEC),through distributed deployment and sinking the data computing center to the edge of users,greatly reduces the delay and energy consumption compared with cloud computing.As a convenient and low-cost aircraft,Unmanned Aerial Vehicle(UAV)can be used as a mobile communication platform to provide effective communication coverage for remote areas.Combining mobile edge computing with UAV can provide flexible location options for MEC servers,which further improves network coverage and service quality.Although UAV-enabled MEC systems provide users with further sinking of computing resources,they are often limited by limited communication,computing,and energy resources.The service duration of UAV can be extended and system latency and energy consumption can be reduced by designing user computation offloading and resource allocation strategies reasonably.Based on the above discussion,this paper studies two typical problems of computation offloading and resource allocation in UAV-enabled MEC system.The specific research contents and contributions are summarized as follows:(1)We research on the multi-user computation offloading problem in UAV-enabled MEC system.Firstly,we built a MEC system model in which UAV acts as a fixed base station to provide computation offloading services to users.On the basis of this system model,the optimization problem of minimizing the weighted sum of system energy consumption and task delay is proposed by considering the duality of offloading variables and uplink bandwidth constraints.To tackle this optimization problem,an Actor-Critic based deep reinforcement learning algorithm is proposed.In this algorithm,the multiple deep neural networks are used to jointly generate computation offloading decisions and the slack computation offloading decisions are quantified by an order preserving quantization with noise method.Simulation results show that the proposed algorithm can converge quickly and effectively,which reduces the system energy consumption and latency compared with other benchmark algorithms.(2)We study the trajectory design and resource allocation problem in UAV-enabled MEC system.Firstly,we built a UAV-enabled MEC system where a UAV acts as a mobile base station to provide data collection and energy transfer services for widely deployed Internet of Things devices.On the basis of this system model,a joint optimization problem on the three objectives of system throughput,total harvest energy and UAV energy consumption during the mission is established.Considering the uncertainty and dynamics of the MEC networks,a deep reinforcement learning algorithm based on recurrent deterministic policy gradient is proposed,which achieves the joint optimization of multiple objectives by reasonably designing the flight trajectory of the UAV and allocating the computation resources.The simulation results demonstrate the convergence and effectiveness of the proposed algorithm,which can significantly improve the system efficiency and can achieve the trade-off between multiple optimization objectives by setting the weighting parameters.(3)We validate the proposed algorithm based on Air Sim simulation platform.Considering that developing and testing MEC algorithms in the real world is an expensive and time-consuming process,Air Sim,a UAV simulation platform based on Unreal Engine,provides physical and visual realistic simulation for the test of mobile edge computing algorithms.Therefore,we build a UAV-assisted mobile edge computing simulation environment based on Air Sim,which verifies the effectiveness of the proposed computation offloading and resource allocation algorithms.