AI-4-SPS: An Artificial Intelligence-Driven Semantic Database for Shale Pore Structure Characterization

Understanding shale pore structure across nano- to mesoscales is essential for reservoir evaluation, which supports effective geological exploration and development. However, this understanding remains constrained by data fragmentation across different methods and by the limited integration of small-angle neutron scattering (SANS), which can improve cross-scale shale pore structure characterization at high resolution and nondestructively. In this work, we developed AI-4-SPS, a comprehensive database system comprising a data set, an artificial intelligence (AI) knowledge extraction module, and an AI query agent. The data set compiles data over 250 shale samples from more than 160 publications. Samples cover major Chinese basins and U.S. basins. The data set includes mineralogy, total organic carbon, and porosity and connectivity data from SANS, mercury intrusion porosimetry, and nuclear magnetic resonance. In experiments, AI-4-SPS can assist in shale SANS experiment design, data interpretation, and data comparison with other pore structure characterization methods. In pore structure analysis, AI-4-SPS can assist with data comparison among multiple methods across different areas. Comparative analysis revealed similar SANS porosities between Chinese and U.S. samples. By integrating SANS and MIP porosity through overlapping nanoconnected pore volumes, we refined shale total porosity estimates and partitioned the pore size distributions. Comparison showed that Chinese shales exhibited higher pore connectivity, which was driven by nanoconnected pores. In contrast, meso-connected pores prevailed in U.S. samples, contributing to greater shale oil productivity. The flexibility of the AI knowledge extraction model enables extraction of any user-defined data from publications, including lithofacies characterization, fluid transport, and fracture characterization.