Research on deep reinforcement learning multi-path routing planning in SDN

Abstract

Based on the research of current SDN traffic scheduling technology, combining the advantages of reinforcement learning in strategy optimization and the characteristics of SDN network resource centralized control, and using the neural network of depth learning to fit the Q-value table, this paper proposes an intelligent multi-path routing planning method based on depth reinforcement learning. Experiments show that the method can use the characteristics of SDN to route the network traffic in multi-path according to the current state information and traffic characteristics of the network; using the advantages of reinforcement learning, it can find multiple forwarding paths for different flows that conform to their traffic characteristics, and improve the utilization rate of the network link bandwidth; and using the neural network of deep learning to fit the Q value in the traditional reinforcement learning algorithm Table.