Dynamic Offloading Optimization Of Edge Computing For B5G

With the development of the fifth-generation mobile communication technology and the popularization of Internet of Things technology and artificial intelligence,the number of smart mobile devices connected to the network has shown an explosive growth trend in recent years,resulting in billions of connections and EB-level network traffic.Additionally,users of these smart devices usually require high-reliability,low-latency communication connections.In this context,distributed mobile edge computing has emerged as a key technology to meet user demands in severe network environments.Since 2006,the development of cloud computing has brought solutions to the shortcomings of mobile devices in computing power,however,in the context of massive connections,compared to centralized cloud computing,distributed mobile edge computing is more suitable to meet the demands of low latency and massive connections,in which content sharing through D2 D communication has become a key solution to alleviate the pressure on the cellular network and meet the low latency requirements of users.However,there are still many issues worthy of study,this article will investigate the optimization of the centralized and the distributed decision-making in the network edge,from the perspectives of the design of the incentive mechanism and the users' transmission decision,respectively.In order to encourage mobile devices to participate in content sharing,this article first studies the design of incentive mechanism,the purpose is to maximize the long-term offloading rate of the network traffic.This article simulates the problem as a market transaction model,selects content providers by scoring and grants corresponding rewards.In order to achieve the effect of continuous incentives under the premise of ensuring the quality of service,the historical performance and current state of the mobile devices are comprehensively considered in the proposed scoring mechanism.Moreover,the subsequent impact on time brought by the introduction of historical performance is analyzed in detail in this article,and the maximization of the offloading rate is formulated as a stochastic dynamic programming,which is solved by an applied deep reinforcement learning algorithm.The simulation results show that the deep reinforcement learning algorithm is suitable for the stochastic dynamic programming problem in this article,and the designed incentive mechanism can effectively and continuously encourage the devices to participate in content sharing under the premise of ensuring the quality of service.The design of the incentive mechanism is from the perspective of the network scheduler to achieve the global optimization of the network.In the transmitting process of the sharing contents,the distributed decision-making on the device side is also worth studying.Taking into account the mutual interference of communications in the network,the distributed decision-making problem of devices is modeled as a non-cooperative dynamic stochastic gaming in this article,and transformed into a lower complexity mean field game problem.For the Hamilton-Jacobi-Bellman equation,the partial differential equation that describes the optimal decision in the gaming,an artificial neural network is designed to apply supervised learning to solve it.The simulation results show that the forward propagation of the network yields relatively low computational complexity,and the training of the neural network converges quickly.The comparison with the Finite Difference Method shows that the network can well fit the equation to solve.The simulation results also show that the decision-making based on the mean field game achieves lower cost with compared to conventional methods.