Research On Computational Offloading Algorithms In Mobile Edge Computing Scenarios

In the era of Io T,Mobile Edge Computing(MEC),a new type of cloud computing architecture,can significantly reduce the computational latency and energy consumption of user devices.A reasonable computation offloading strategy can fully utilize the server resources and improve the user experience.In this paper,the optimization algorithm of computation offloading strategy is studied based on two different MEC scenarios,and the main research contents are as follows.(1)For the multi-user single-server scenario,a system model is built with the optimization goal of minimizing the system cost in the mode of full offloading of tasks,and an improved DL-GA policy optimization algorithm is proposed to optimize the system cost in view of the drawback of the long execution time of Genetic Algorithm(GA).By introducing the CBAM mechanism to optimize the neural network model of DL-GA,the characterization ability of the neural network is effectively improved.Simulation results show that the system cost optimized by the improved DL-GA algorithm is reduced by about 2%～3.7% on average compared to the DL-GA algorithm with similar execution speed.The system cost optimized based on the GA algorithm is reduced by about 4.4%～7.8% on average compared to the improved DL-GA algorithm,however,the execution speed of the improved DL-GA algorithm is 2400～2900 times faster compared to the GA algorithm,which verifies the effectiveness of the improved DL-GA algorithm.(2)For the multi-user multi-server scenario,firstly,task offload fairness is considered as an optimization index in the mode where tasks are partially offloaded,which consists of two components: task priority and the balance of offload quantity among devices.The quality of service is used as a weighted sum of system cost and task offload fairness,and the system model is built by maximizing the quality of service.A DQN policy optimization algorithm based on deep reinforcement learning is designed and compared with the improved DL-GA algorithm.Simulation results show that the system cost based on the improved DL-GA algorithm optimization decreases by 9.4% to 11.7% on average compared to the DQN algorithm,but the optimization based on the DQN algorithm is about 5.7% to 9% higher than the DL-GA algorithm on average in terms of task offloading fairness.