Routing and Spectrum Assignment in Elastic Optical Networks through a Hybrid Approach based on K-Shortest Paths and Q-Learning

Reinforcement learning techniques have been successfully applied to the routing and spectrum assignment problem in elastic optical networks. These techniques offer a promising approach by enabling agents to make sequential decisions based on network conditions and traffic demands. However, the literature reported that heuristics are more efficient than reinforcement learning approaches in low-traffic scenarios. Consequently, this paper explores the advantages of combining the strengths of both approaches to achieve an efficient algorithm in all loading scenarios. This study explores the performances of Q-learning for medium and high traffic loads and the K-shortest path algorithm for low traffic loads. The proposed hybrid approach leads to improved resource utilization, latency reduction, and overall network performance under dynamic traffic load. Numerical simulations were performed on a real network topology in which the proposed hybrid approach was studied against the state-of-the-art algorithms. The results indicate that the proposed approach shows promise in achieving better or equal performance at all traffic loads compared to non-hybrid approaches.