Network Reliability Enhancement Using Machine Learning and Metaheuristics

SUMMARY & CONCLUSIONSThis paper presents a hybrid framework that combines Graph Neural Networks (GNNs) and metaheuristic optimization to enhance the all-terminal reliability of a dynamic infrastructure network. A Principal Neighborhood Aggregation (PNA)-based GNN is trained as a surrogate model to efficiently approximate network reliability, significantly reducing the computational burden compared to those exact methods. This surrogate is embedded within metaheuristic algorithms—Particle Swarm Optimization (PSO), Genetic Algorithm (GA), and Grey Wolf Optimization (GWO)—to optimize maintenance and improvement actions under budget constraints. Experimental results show that PSO achieves the best performance in maximizing network reliability with strong stability, while GA and GWO offer varying levels of effectiveness. The proposed approach demonstrates scalability and adaptability to large-scale networks, offering a practical and data-driven solution to overcome technical challenges in reliability enhancement and maintenance planning.