Graph Attention Network Enhanced Deep Reinforcement Learning Framework for Routing, Modulation, and Spectrum Allocation in EONs

We propose a graph attention network (GAT) en-hanced deep reinforcement learning framework for routing, modulation and spectrum allocation (RMSA) in elastic optical networks. GAT is able to learn the relationships between links, which is important for RMSA due to the spectrum continuity constraints. Simulation results show that our proposed method outperforms state-of-the-art methods.