On Board-Level Failure Localization in Optical Transport Networks Using Graph Neural Network

This paper investigates a novel framework for board-level failure localization in the Optical Transport Networks (OTN), dubbed Board-Alarm Propagation Tree based Failure Localization (BAPT-FL). Foremost, a collection of functional graphs (FGs) is garnered by iteratively tagging each board in the network topology, serving as the ground of the proposed framework. Concretely, BAPT-FL is designed to build a range of BAPTs by correlating the tagged boards and alarms involved in the FGs, where each BAPT deems a failed board and its corre-lated alarms as the root and leaves, respectively. To evaluate the edge weights of potential BAPTs induced by FGs, a graph neural network (GNN) with the graph transformer operator is employed as an edge classifier, which characterizes each vertex/edge from diverse dimensions including time, traffic distribution, network topology, and board/alarm attributes. Subsequently, we frame an integer linear programming (ILP) problem to construct the best possible BAPT(s). Extensive case studies are conducted to showcase BAPT-FL's advantage over its counterparts in terms of the metrics assessing the identified failed boards/root alarms. We also delve into its performance in volatile environmental variations such as diverse failure scenarios, network topologies, traffic distributions, and noise alarms.