Deep Reinforcement Learning for Resource Allocation in Multi-Band Optical Networks

Routing and Spectrum Assignment (RSA) is key to an efficient resource usage in optical networks. Although this problem is known to be complex, an even more complex version arises when considering multi-band (MB) optical networks, where the spectrum-dependency of performance becomes significantly more pronounced. This paper proposes a Deep Reinforcement Learning (DRL)-based strategy for RSA in MB optical networks leveraging the GNPy library for accurate estimation of optical performance. Simulation results show that DRL-RSA reduces blocking by up to 80% when comparing to state-of-the-art RSA strategies.