Modeling and upgrade of disaster-resilient interdependent networks using machine learning

Abstract

Recent global emergencies emphasize the critical role of reliable communication networks. As dependence on critical infrastructures grows, the focus shifts from isolated failures to designing networks capable of withstanding disasters, taking into account their interdependence with infrastructures like the power grid. This paper investigates the problem of the disaster resilient upgrade of interdependent networks, focusing on enhancing network resilience during emergencies and ensuring a service-level agreement. We analyze how the interdependency between the networks affects the disaster resilience and propose heuristic methods for network operators to improve resilience against disasters. Furthermore, to address the challenge of hidden interdependencies, we present a novel approach using graph neural networks for predicting interdependency between networks based on historical data of failures. Using simulations with real networks and earthquake data, we demonstrate that limiting the number of interdependent edges per node significantly affects resilience. We show that if sufficient data is available graph neural networks can learn the connection between failures and interdependencies, and capable of predicting interdependencies. Additionally, we show that selecting appropriate upgrade methods can reduce network upgrade costs by up to 20%.