A deep neural network model for link failure identification in multi-path ROADM based networks

Critical telecommunications infrastructures, such as the modern Datacenters and 5G networks, require a reconfigurable optical add/drop multiplexer (ROADM) optical network (ON) backbone with high availability. In such ONs, the reduction of the time to identify and repair failures is very important. As this process involves the analysis of a large amount of data, machine learning techniques become essential to identify network faults. In this work, we use a deep neural network (DNN) model to identify the presence of failures in a complex multi-route ON. Preliminary evaluation tests achieved 85% accuracy in the identification of failures in a given link using DNNs.