Smart Resource Utilization in Elastic Optical Network Using Generative AI

Efficient Routing and Spectrum Assignment (RMSA) in Elastic Optical Networks (EONs) is essential for addressing dynamic and bursty traffic demands. This paper introduces an innovative framework that integrates the Deep Reinforcement Learning (DRL) technique, Proximal Policy Optimization (PPO), with Generative Adversarial Networks (GAN) to improve decision-making processes for RMSA. The GAN is utilized to predict future network states, including traffic demands and spectrum availability, which are subsequently incorporated into the state space of the PPO agent. Furthermore, integrating the Large Language Model (LLM) facilitates proactive and adaptive resource allocation by guiding the reward engineering process. Simulation results indicate that the GAN-augmented PPO framework, incorporated with LLM, substantially decreases blocking probability while enhancing the throughput compared to the traditional method.