Spectrum Adaptive Awareness Routing and Spectrum Allocation Based on Reinforcement Learning

We propose a spectrum adaptive awareness routing and spectrum allocation (SAA-RSA) scheme based on reinforcement learning in elastic optical networks. This scheme combines spectrum fragment awareness and reinforcement learning to realize adaptive spectrum allocation. Based on the proposed scheme, the action space of RL is optimized, thus improving the performance of network and reducing traffic blocking probability. The simulation results show that when the network load is 300 Erlang, compared with traditional KSP-FF method and reinforcement learning method without considering spectrum fragment, the scheme can reduce blocking probability by 27.56% and 12.15% respectively.