DGL-Routing: One Routing Optimization Model Based on Deep Graph Learning

Traditional routing scheme cannot be applied in the dynamic and complex network environment directly. Thus resulting in degraded performance. This paper proposes one Deep Graph Learning (DGL) algorithm based on Graph Convolution Network (GCN) and Actor-Critic architecture. The GCN is selected as the policy network in order to update the link weight of the whole network. The GCN is trained by the critic network. The network traffic is allocated proportionally according to the total weight of the reachable path. Finally, the GCN realizes the routing optimization of the global network. Simulation results demonstrate that the proposed scheme is superior to the comparison scheme in terms of network average end-to-end delay, packet loss rate, and throughput. The proposed model can maintain stability and achieves the best optimization performance in different topology structures as long as it is trained once.