Aplicação de técnicas de aprendizado de máquina para planejamento e gerenciamento de redes ópticas elásticas

Abstract

Elastic Optical Networks (EONs) have emerged as an innovative response to traditional optical networks, bringing new operational concepts that improve flexibility and resource efficiency. A recurring problem in EONs is Routing and Spectrum Allocation (RSA), which seeks to define a route for each request and allocate an appropriate number of slots according to the required demand, using the minimum possible spectrum. This work presents supervised machine learning techniques for the virtualization design with protection in EONs, aiming to predict the total number of spectrum slots needed to support all traffic demands. Focusing on Virtual Optical Networks (VONs) subjected to specific protection, the application of machine learning techniques, specifically Multilayer Perceptron (MLP) and Support Vector Regression (SVR), is investigated to solve the link capacity problem in EONs with virtualization faster than traditional Integer Linear Programming (ILP) formulations, while maintaining results close to optimal. The performance of the models was evaluated through statistical metrics, training time, and inference. The results showed that the proposed method is effective in predicting the number of slots needed in the physical substrate subjected to various VONs.