A Routing Optimization Policy Using Graph Convolution Deep Reinforcement Learning

The diversification of network traffic has brought about more serious quality of service (QoS) issues. Existing QoS optimization methods based on reinforcement learning and neural networks are unable to characterize the structural features of network topology, resulting in suboptimal optimization performance. In this paper, a routing optimization method combining Graph Convolutional Neural Network (GCN) and Deep Deterministic Policy Gradient (DDPG) is proposed to address QoS issues in Software Defined Network (SDN). The GCN module analyzes dynamic network status and extracts network topology structure information, which is used by the DDPG module to make routing decisions. The proposed strategy outperforms OSPF algorithm, DRL-TE strategy, and DDPG routing algorithm in terms of optimizing average end-to-end delay and packet loss rate.