Research On Relay-Based Cooperative Task Offloading Method In Mobile Edge Computing

In recent years,to meet the higher requirements of various new application technologies（such as augmented reality and metaverse）on delay,energy consumption and energy efficiency,mobile edge computing（MEC）provides a new computing paradigm.With the increase in the number of edge servers and the expansion of service scope,there are often idle computing re-sources that are not fully utilized,and also exists a problem of server computing overload.Com-pared with traditional MEC,relay-assisted MEC can effectively improve system performance by using idle nodes to assist computing and forwarding.Thus,how to design a reasonable task offloading strategy in the relay-assisted MEC system has great research significance.This pa-per focuses on the key difficult problems in the relay-assisted MEC,such as tasks randomness,resource constraints and information asymmetry,etc.,and uses Lyapunov optimization,deep learning and contract theory to solve the problem of task offloading in relay-assisted MEC.The specific research contents are as follows:（1）To address the problem of task randomness and resources constraints in the relay-assisted MEC system,relay selection and task offloading decisions are considered comprehen-sively to minimize the long-term average energy consumption of the relay-assisted MEC system under the constraints of task buffer queues stability.The problem is formulated as a mixed inte-ger nonlinear stochastic optimization problem,and is solved by decomposing into two stages:relay selection and task offloading decision to solve.In the relay selection stage,the relay node is determined by introducing the weight parameter V₁to minimize the weighted sum of user transmission energy consumption and the length of the relay buffer queue.In the task of-floading decision stage,the stochastic optimization problem is transformed into a deterministic optimization problem by Lyapunov method.Under the condition of keeping the task buffer queue stable,the theoretical expressions of the optimal relay calculation frequency,optimal re-lay transmission power and optimal remote calculation frequency can be obtained.Finally,the simulation results show that the energy optimization strategy can effectively reduce the long-term average energy consumption of the system under the constraints of buffer queue stability,and obtain a result converging to the optimal solution of exhaustive search.（2）To solve the problem of time-varying environment and information asymmetry of the unmanned aerial vehicle（UAV）-assisted MEC system,a deep reinforcement learning-based contract incentive（DRLCI）task offloading strategy is proposed,with the goal to maximize the long-term utility of all hot spots（HSs）.This is a joint optimization problem of offloading decision and contract design.Due to the stochastic user demands and time-varying wireless channel conditions,it is necessary to jointly optimize offloading decision and contract design over time to achieve maximum performance.DRLCI solves this problem by two steps.First,a deep Q-network（DQN）algorithm is used to obtain offloading decision,the Double-DQN framework is used to ensure a more stable learning process,and the Dropout layer is used to prevent the neural network from overfitting during forward propagation.Secondly,in the case of information asymmetry,a contract-based incentive mechanism is designed to attract different types of UAVs to participate in resource sharing.The optimization problem can be simplified by analyzing the sufficient and necessary conditions of feasible contracts,and the Lagrange multiplier method is used to approximate the optimal solution.Finally,simulation experiments verify the effectiveness of the contract by analyzing incentive compatibility（IC）and incentive rationality（IR）conditions.In addition,the proposed DRLCI strategy in this paper can effectively adapt to time-varying environments and achieve better convergence speed than natural DQN and Double-DQN.