Securing CO2 storage in fractured reservoirs: Intelligent characterization of critical natural fracture pathways

Accurate characterization of natural fractures (NF) in depleted reservoirs is critical for mitigating CO₂ leakage risks during Carbon Capture, Utilization and Storage (CCUS) deployment. However, conventional methods fail to address the core challenge of reliably segmenting fragmented fracture networks in Electrical Imaging Logging (EIL) images, leading to significant uncertainties in storage integrity assessment. To overcome this limitation, we propose a novel dual-core intelligent framework integrating prior knowledge-guided graph representation and graph contrastive learning. Specifically, NF features are extracted via pixel-node transformed path-graphs that explicitly encode spatial attributes (e.g., fracture density, azimuth), followed by knowledge-based multi-threshold denoising to eliminate non-conductive artifacts. Crucially, a graph contrastive learning model is introduced to analyze morphological dependencies among fracture nodes, enabling robust matching of fragmented segments under subtle variations—a capability unattainable by traditional machine learning methods. Validated with field data from Southwest China, our approach achieves a breakthrough accuracy of 96.37% in fracture identification, significantly outperforming existing techniques. This study contributes the scalable solution for reconstructing complete NF networks from fragmented EIL images, providing reliable characterization of critical CO₂ leakage pathways. By ensuring reliable reservoir potential assessment, our method directly enhances storage security and advances the safe deployment of carbon sequestration in renewable energy transition initiatives.