基于强化学习的 SDN 路由规划算法的实现

raditional routing protocols use limited information to make routing decisions, which can lead to slow adaptation to traffic variability and limited support for quality of service (QoS) requirements for applications. The related research of Software Defined Networking (SDN) provides the possible deployment of intelligent routing algorithms, allowing the routing layer to complete more and more complex tasks.Although the centralized control of SDN provides a control framework for realizing centralized network optimization, the adaptive routing algorithm of different traffic loads requires a complex model, so how to realize load balancing in SDN network architecture through reasonable flow scheduling algorithm, improve the service quality of the application is still in a preliminary stage. In recent years, the artificial intelligence technology based on reinforcement learning has been successfully applied to a wide range of complex control and optimization problems, especially the classic tabular reinforcement learning, also known as Q-learning. The gradual improvement of this learning method has provided a technical basis for the application of reinforcement learning technology to the routing field. Combined with the advantages of both, this paper proposes a SDN routing planning method (Q-SDN) based on reinforcement learning, which is found to have better adaptive ability and can meet the diversified optimization needs of various performance evaluation indicators.