Busyness level-based deep reinforcement learning method for routing, modulation, and spectrum assignment of elastic optical networks

Deep reinforcement learning (DRL) has been introduced in routing, modulation and spectrum assignment (RMSA) of the elastic optical networks. Since the DRL agent’s learning is based on the state it observes and the reward it receives, key information should be embedded in the state and the reward. In previous studies, the observed and feedback information is limited. In this paper, we propose a busyness level-based DRL method for the RMSA of the elastic optical networks. Since the busyness of the links or transmission paths highly affects the performance, we believe the busyness information should be perceived by the agent to learn a good RMSA policy. Specifically, we define two indicators to quantify busyness level, and then combine these two indicators into the design of reward and state. Simulation results show that our approach works better than the case that busyness is not