Strategic scenario planning for hydrogen supply chains via graph convolutional networks: overcoming mid-stream infrastructure bottlenecks in energy transitions

: Hydrogen transitions require a combination of change in policy, production, distribution, and adoption, but current scenario consistency assessment only checks for trend coherence without incorporating the role of the relationships among the stakeholders in shaping those emerging pathways. This study introduces a Graph Convolutional Network-based Scenario Planning (GCN-SP) framework that extends scenario-consistency assessment by linking actor-node engagement with emerging trends to graph-based representations of representative hydrogen scenarios. Multi-driver interaction effects are transformed into strategic scenario graphs, clustered using k-means to derive representative configurations from cluster centroids, and classified using a graph convolutional network. The optimal model achieves weighted precision and recall exceeding 0.78 under both cross-validation and hold-out validation and outperforms two multilayer perceptron baselines. Three distinct structural configurations emerge. Coordinated: aligned actors across government, production, distribution, and adoption form a mutually reinforcing structure supporting system integration. Divergent: production and international connectivity advance while the producer becomes disconnected from the domestic network, creating an export-oriented pathway. Constrained: supply-side ambition and transport expansion outpace downstream adoption, leaving the value chain incomplete. Across scenarios, distribution emerges as a structurally important link whose connectivity determines whether production translates into integration. These results show that similar stakeholder activity can generate fundamentally different outcomes depending on interaction configuration. GCN-SP therefore enables structural distinction between coordinated integration, export-oriented divergence, and supply-led constraint.