Intelligent P-cycle provisioning in elastic optical networks via energy-aware deep reinforcement learning

Abstract To accurately assess the service accessibility of networks using preset cycles (p-cycles), we construct a theoretical model in this research. Compared to earlier work, our model is able to predict service availability more accurately by taking into account the dependency across protection domains as well as all instances where a lightpath is accessible. We next provide an approach for dynamic elastic optical networks (EONs) that is directed toward service availability throughout the configuration cycle, based on the results of the availability study. In elastic optical networks, a hybrid strategy combining deep reinforcement learning (DRL) and an improved genetic algorithm (IGA) is introduced to mitigate spectrum fragmentation and excessive resource redundancy associated with traditional pre-configured cycle (P-cycle) protection. The DRL agent learns to identify optimal base cycles, while the IGA refines their combination to achieve efficient, network-wide protection coverage. To further reduce spectrum fragmentation, a link-based spectrum partitioning approach integrated with P-cycle configuration is proposed, enabling better organization of spectral resources. Additionally, a routing mechanism is designed to select working paths either by prioritizing minimal traffic load or by applying a first-fit strategy, considering both spectrum utilization and path length to enhance overall efficiency. The results of the simulation demonstrate that the suggested DRL may enhance spectrum resource use while decreasing the likelihood of bandwidth blockage. Two benchmark-cycle design techniques that utilize the protection effectiveness of -cycles are compared with the suggested approach in the simulations. The proposed method can effectively enhance service accessibility in EONs with near minimal alteration to blocking probability as indicated in the simulation outcomes.