The Application Of Deep Reinforcement Learning In Mobile Edge Computing

Mobile Edge Computing refers to providing certain computing capabilities for near mobile phones and mobile terminals,that is,giving some edge computing capabilities to edge base stations so that they have the ability to assist users in intelligently processing services.Mobile Edge Computing "sinks" network services into the wireless access network,so it has three major advantages: lower latency,effectively suppressing network congestion,more network information and network control functions can be open to the developers.Deep Reinforcement Learning is a learning method that combines Deep Learning and Reinforcement Learning.Deep Learning is used to provide a learning mechanism,while Reinforcement Learning provides the goal of Deep Learning.It combines the deep understanding of Deep Learning in problem perception and the exploration and decision-making ability of Reinforcement Learning,which makes it able to solve more complex problems in real scenarios.Because the application requirements in Mobile Edge Computing are in line with the relevant characteristics of Deep Reinforcement Learning,that is,the complexity of the problem and the need to get a solution to the problem in time,using Deep Reinforcement Learning to solve the application problems has become a direction that can be explored.This article mainly focuses on Deep Reinforcement Learning methods,such as DQN,A3 C,etc.,to solve the online cache strategy problem,cache replacement problem,and online DASH resource allocation problem in Mobile Edge Computing.It mainly includes the following research contents:1)Aiming at the online cache strategy problem,the DQN method is adopted to solve the corresponding policy decision-making problem,which can achieve rapid convergence in stable scenarios and timely adaptation to changes in changing scenarios,while achieving good results.2)Aiming at the problem of online cache replacement,the A3 C framework and a new reward function mechanism can be used to achieve a balance between Cache Hit Rate indicators and cost.3)Aiming at the problem of online DASH resource allocation,A2 C and DDPG methods are used to achieve multi-user resource allocation without stuttering,which improves user experience without causing stuttering problems,while also taking intoaccount some fairness.The above work have passed a certain amount of simulation tests and performance comparison tests with corresponding comparison algorithms.At the same time,the experimental verification has been performed in multiple scenarios.The related Deep Reinforcement Learning methods can be based on low-latency operation.It has achieved improvement in corresponding indicators,and has better performance than traditional methods,thereby solving related problems in cache and DASH services.