Research On Offloading Strategy Of Mobile Edge Computing Tasks Based On Reinforcement Learning

With the rapid development of wireless communication technology,Internet of Things technology and 5G technology,mobile devices are rapidly popularized,and data traffic has increased sharply.Some emerging applications such as online games,artificial intelligence and virtual reality have relatively high requirements for time delay and require a lot of computing resources.However,the computing power and battery capacity of the mobile terminal are limited,and running these applications will bring high computing delay and increase the energy consumption of the mobile terminal.The service mode based on cloud computing center is often difficult to meet the needs of real-time applications due to the transmission delay caused by the long transmission distance.As a result,the Mobile Edge Computing(MEC)paradigm emerged.By sinking services to the edge of the network,the network transmission delay can be reduced and the demand for low-latency services can be met.Among them,computational offloading is one of the key technologies in MEC.However,compared with the increasing computing needs of mobile users,the limitation of MEC server computing resources is becoming more obvious due to constraints such as hardware costs.Therefore,how to design a reasonable computing offloading strategy to meet the needs of mobile users is a huge challenge.Existing traditional algorithms such as Game Theory(Game Theory),Genetic Algorithm(GA)and Ant Colony Optimization(ACO),etc.Although these heuristic algorithms have good performance in their respective goals,such as reducing the user's delay cost or energy consumption cost.However,the time slot interval divided by the MEC system is very small,and the traditional algorithm requires a large number of iterations to obtain an optimized solution,so the traditional optimization algorithm is not suitable for high real-time MEC systems.Reinforcement learning is a trial-and-error learning method.The optimization goal is achieved through a trial-and-reward mechanism,and the trained model is deployed on the MEC server,which can effectively improve the response speed of the MEC system.Based on reinforcement learning and deep learning,this paper conducts an in-depth study on the problem of computing offloading in edge computing.The main tasks are as follows:(1)Research the problem of computing offloading based on task cache in a single-cell scenario.A joint offloading and task caching strategy(Joint Offloading and Caching,JORC)is proposed.The JORC mechanism comprehensively considers the schemes of task offloading,communication and bandwidth resource allocation.For task caching,the value of caching is defined based on multiple attributes of the task.First,model the offloading,resource allocation and task caching models.Secondly,the computational offloading and resource allocation problems are formalized and modeled as a Markov model.The evaluation index is to minimize the task completion delay and the weighted sum of energy consumption,and finally solved by the QLearning-based algorithm.The simulation comparison of this algorithm with other benchmark schemes proves that this algorithm can effectively reduce time delay and energy consumption.(2)In the scenario of multi-user and multi-base stations,a master-slave MEC cooperative computing offloading and resource allocation strategy is designed,combined with task priority and time delay constraints,and a scheduling algorithm based on task priority is proposed.First of all,virtualize the network and establish a communication model,offloading model,collaboration model,and task priority model based on this.Secondly,an offloading scheduling algorithm based on deep reinforcement learning is proposed.In the algorithm,the user's completion delay in the system is designed as the reward and loss of deep reinforcement learning,and through neural network training,the algorithm learns in the exploration to obtain the highest reward offload and resource allocation method,ensuring the MEC server of the master and slave cell Load balancing minimizes the weighted sum of all user delays and energy consumption,and at the same time improves the calculation success rate of the task.Finally,through simulation and comparison of multiple benchmark algorithms and training models,it is proved that the proposed model can obtain higher system benefits.