Research On Computation Offloading Algorithm Based On Game Theory And Reinforcement Learning In Mobile Edge Computing

At present,with the development of the Internet of Things and the Internet of Everything,more and more mobile devices are appearing in people's lives,which leads to the number of network edge devices exploding.Therefore,traditional cloud computing cannot support large-scale computing tasks.Mobile edge computing provides cloud computing capability at the edge of wireless access network to solve the problem of limited resources of mobile devices.In order to use the services,which is provided by the edge network,how to offload the computing tasks to the edge server and how to make efficient and reasonable computation offloading decisions have become the main research direction of mobile edge computing.In this paper,the computation offloading algorithm is studied,and the main work is summarized as follows:1.For the mobile edge computing,the computation offloading depends on the wireless transmission channels.The process of multi-user competing for the channel is a process of game.The multi-user computation offloading problem is transformed into a game problem.The game algorithm of multi-user computation offloading is proposed.The simulation results show that the algorithm can achieve Nash equilibrium in a finite number of decision slots.2.According to the system model of multi-user computation offloading problem,three elements of reinforcement learning are formulated.The Nash equilibrium of multi-user computation offloading game is added to the exploration mechanism of reinforcement learning.Based on Nash equilibrium,the Q-learning algorithm and the deep Q network algorithm are proposed.By setting the system parameters and neural network parameters to perform simulation,and the results show that the proposed algorithm converges faster and reduces the overhead of the system by 16.8% compared with other algorithms.3.A method of computing resource pricing is proposed to allocate the limited computing resources of mobile edge server.Mobile users purchase the computing resources of mobile edge server to computation offloading,and mobile edge server gains profits by charging mobile users.Therefore,the interaction process is constructed as a Stackelberg game,and a two-layer deep reinforcement learning algorithm is proposed.The simulation results show that the Stackelberg equilibrium can be reached in a finite number of games.4.According to the different utility pursued by mobile edge server and mobile user,two optimization functions(e.g.,maximizing the revenue of mobile edge server and minimizing the computing overhead of mobile users)are formulated.By analyzing the difference between mobile edge server and mobile user's action space,the corresponding reinforcement learning algorithms are proposed,in which the mobile edge server considers two pricing methods:uniform pricing and differentiated pricing.The simulation results show that the proposed algorithm can increase the revenue of mobile edge server by 29.8% and reduce the computing overhead of mobile users by 12.9%.