Machine Learning Assisted Network Resilience Design

This paper focuses on network resilience-based design, which involves carrying out a priori analysis during the design phase to optimize the topology of the current network to increase its resilience against disruptive events. However, solving this problem is computationally expensive as it requires searching through numerous topology structures and conducting multiple resilience analyses under different network topologies, which can become unaffordable for large-scale networks. To address this issue, we propose a new graph neural network that incorporates response flow characteristics to capture additional network features. This neural network is designed to deal with complex problems with high dimensions and nonlinear characterization, and it is integrated into an adaptive framework that combines it with a probabilistic solution discovery algorithm to solve network resilience design problems accurately and efficiently.