Matheuristic with machine-learning-based prediction for software-defined mobile metro-core networks

In general, humans follow a routine with highly predictable daily movements. For instance, we commute from home to work on a daily basis and visit a selected set of places for commercial and recreational purposes during the nights and weekends. The use of mobile phones increases when commuting via public transportation, during lunch breaks, and at night. Such regular behavior creates predictable spatiotemporal fluctuations of traffic patterns. In this paper, we introduce a matheuristic for dynamic optical routing, which can be implemented as an application into a software-defined mobile carrier network. We use machine learning to predict tidal traffic variations in a mobile metro-core network, which allows us to solve off-line mixed-integer linear programming instances of an optical routing (and wavelength) assignment optimization problem. The optimal results are used to favor near-optimal on-line routing decisions. Results demonstrate the effectiveness of our on-line methodology, with results that match almost perfectly the behavior of a network that performs optical routing reconfiguration with a perfect, oracle-like traffic prediction and the solution of an optimization problem.