Research On Joint Allocation Algorithm Of Computation And Communication Resources Based On Reinforcement Learning In Mobile Edge Computing

The era of the Internet of Everything has arrived,and the rapid growth in the number of terminal devices has made the communication network and the Internet constantly under tremendous pressure.The limitations of traditional cloud computing are becoming more and more obvious.The advent of the 5G era has also brought about the booming of emerging technologies such as Mobile Edge Computing(MEC).Mobile Edge Computing has the characteristics of proximity,low latency,high bandwidth and location awareness,which can greatly alleviate the network congestion problems such as the rapid increase of the number of network devices,the explosion of mobile traffic and the excessive proportion of network data retransmission.However,the MEC server deployed in the mobile edge computing system still has to face a large number of data processing requests of User Equipment(UE),and still has high requirements for the allocation and scheduling capabilities of computation and communication resources.This paper proposes an efficient joint allocation algorithm for computation and communication resources based on reinforcement learning for mobile edge computing systems.In order to maximize the number of user equipment benefiting from mobile edge computing and to minimize the average overhead of user equipment,we have jointly optimized data task offloading and power control strategies for each user equipment.The resource joint allocation algorithm with power control is more prominent in reducing the average overhead of user equipment.However,finding the optimal solution of the two optimization objectives proposed in this paper is still a nondeterministic polynomial hard(NP-hard)problem.So,this paper establishes a Markov Decision Process(MDP)model and uses Reinforcement Learning(RL)algorithm to solve it.This paper uses the synchronization strategy control algorithm Sarsa and the asynchronous strategy control algorithm Q-Learning in reinforcement learning to compare with the traditional resource allocation algorithm based on Received Signal Strength(RSS).Through the experimental simulation results,we can see that the proposed joint allocation algorithm of computation and communication resource based on reinforcement learning can achieve approximate optimal performance compared with the traditional exhaustive algorithm.And compared with the RSS-based resource allocation algorithm,better performance is achieved not only from the perspective of the entire system(more terminal equipment benefiting from mobile edge computing)but also from the perspective of individual terminal user equipment(less overall overhead).