Dynamic Control of Transparent Optical Networks with Adaptive State-Value Assessment Enabled by Reinforcement Learning

For efficient and dynamic path operations in transparent optical networks, routing and wavelength assignment (RWA) must be optimized in terms of not only link-resource utilization but also traffic distribution. In this paper, we propose a reinforcement-learning-based RWA algorithm that maximizes the number of paths to be accommodated to a network with pre-training using estimated traffic distributions. Numerical experiments elucidate that the number of paths accommodated increases by up to 9.1%.