| With the booming development of technologies such as Internet of Things(Io T)and Artificial Intelligence(AI),an intelligent society where everything is sensed and connected is coming.On the one hand,due to the rapid increase in the volume of heterogeneous data and the complexity of model training,traditional centralized computing can no longer meet the computing needs of the smart society.On the other hand,the emergence of edge computing and 5G technology makes people start to utilize the resources at the edge side of the network.The computing resources is gradually evolving towards distributed and networked.The computing-power network is proposed as a new paradigm for resource integration.The resources such as computation and storage,which belong to the cloud,the edge and the end,are integrated through efficient network.Then,the unified services are provided outward in the form of resource pools.According to the business characteristics,users can flexibly choose the appropriate computing-power resources and pay on demand.However,most of the current work focuses on multi-level collaborative framework of computing and scheduling of resources.There is a lack of in-depth research on the trading of computing-power resources and the balance of benefits..This thesis constructs a market framework of computing-power network for trading computing-power resources.The respective benefit functions of participants such as users,computing-power provider and computing-power network service provider are analyzed,so as to encourage players to actively participate in the construction of computing-power network.In particular,the network congestion cost is added to the user’s benefit model,considering the impact of network congestion effects on the service quality of computing-power.Then,since the participants are rational and selfish to maximize their own profits,the resource trading is modeled as a three-stage Stackelberg game with sequential decisions to balance the benefits of all players.The existence of Nash equilibrium is proved theoretically.Then,this thesis designs the dynamic-game reinforcement learning(DG-RL)algorithm with adaptive learning rate to solve the optimal strategies under the environment of incomplete information game.The experimental results show that the DG-RL algorithm can converge quickly and the algorithm policy solution is almost close to that in the optimal algorithm,which verifies the effectiveness of the market framework and the excellent performance of the proposed algorithm.In addition,this thesis also studies the impact of changes in the number of users,congestion effect coefficient and data transmission price on the computing-power network market. |
