Machine learning models for estimating quality of transmission in DWDM networks

It is estimated that 5G and the Internet of Things (IoT) will impact traffic, both in volume and dynamicity, at unprecedented rates. Thus, to cost-efficiently accommodate these challenging requirements, optical networks must become more responsive to changes impacting the traffic and network state as well as operate more closely to optimality. In this context, knowledge-defined networking (KDN) promises to play a paramount role in improving network flexibility and automation. KDN is a solution that introduces reasoning processes and machine learning techniques into the control plane of the network, enabling it to operate autonomously and faster. One of the key aspects in this environment is the accurate validation of lightpaths. Accurate lightpath validation demands running computationally intensive performance models, which can be time-consuming and impact time-critical applications (e.g., optical channel restoration). This work evaluates the effectiveness of various machine learning models when used to predict the quality of transmission (QoT) of an unestablished lightpath, speeding up the process of lightpath provisioning. Three network scenarios to efficiently generate the knowledge database used to train the models are proposed as well as an overview of the mostused machine learning models. The considered models are: K-nearest neighbors, logistic regression, support vector machines, and artificial neural networks. Results showthat, in general, allmachine learningmodels are able to correctly predict the QoTofmore than 90% of the lightpaths. However, the artificial neural networks (ANN) model is the model presenting better generalization, being able to correctly predict the QoT of almost 99.9% of the lightpaths. Moreover, ANN is able to estimate the residual margin of a lightpath with an average error of only 0.4 dB.