Machine learning in reservoir rocks characterization: integrating seismic data resolution-enhancement for seismic facies classification

Abstract

Amid increasing interest in the dual enhanced oil recovery (EOR) and carbon geological sequestration (CGS) programs, improved static reservoir models emerge as a requirement for well-guided decision-making pertaining to the design of injector-producer well-drilling patterns. To this end, this study utilizes unsupervised machine learning approach leveraged with seismic resolution data preconditioning and spectral analysis to evaluate seismic facies based on machine learning models of clustering in multi-attributes space of the Mississippian carbonates of Kansas. The study provides a benchmark for understanding seismic facies distribution and implications for reservoir aspects pertaining to Enhanced Oil Recovery (EOR) and/or Carbon Geological Sequestration (CGS) programs, especially when encountering sparse well-logs control. A 3D seismic reflection P-wave data and a suite of well-logs and drilling reports constitute the data used for seismic facies based on seismic attributes input to machine learning hierarchical analysis and K-means clustering models. The results of seismic facies, six facies clusters, are analyzed in integration with the target-interval estimated mineralogy (Calcite-Dolomite-Quartz) and a predicted reservoir porosity. The study unravels the nature of the seismic (litho)facies interplay with porosity, sheds light on interpreting unsupervised machine learning classification of Kansas Mississippian carbonates at multi-resolution levels, and paves the way for an improved static model to enable effective CO2-EOR and geosequestration decision making.