Graph Neural Networks Model to Generate Transportation Test Networks

Abstract

Transportation researchers often rely on test networks to evaluate the performance of traffic assignment algorithms, analyze resilience and criticality, study emission productions, investigate transportation economics, and address various other challenges. Existing test networks have limited diversity and might not fully represent the topological features of real-world transportation networks. To address this issue, we proposed a novel approach that leverages real-life road networks extracted from OpenStreetMap to generate synthetic test networks with desired topological properties. We developed a graph neural network model that learns from the extracted networks’ topological patterns and predicts binary adjacency matrices for the generated networks. The proposed model creates synthetic networks resembling real-life networks, demonstrating its effectiveness in generating realistic test networks. The resulting synthetic networks offer a diverse and comprehensive set of test networks for various transportation studies, enabling improved simulations, analyses, and decision making for transportation researchers and practitioners. We have made our source code publicly available for collaboration and further enhancements in the transportation domain.