Application of a semi-supervised self-organizing map algorithm for seismic facies classification using inversion attributes

This study introduces a semi-supervised Self-Organizing Map (SOM) approach for seismic facies classification, effectively overcoming limitations of traditional methods. Initially, unsupervised SOM clustering organizes seismic inversion attributes Impedance (IP), S-Impedance (IS), and Poisson’s Ratio (PR) into a meaningful topological map. Subsequently, labeled well-log data refines these clusters, significantly enhancing geological interpretability. Results demonstrate that even limited training data yield accurate, generalized facies predictions due to SOM’s inherent robustness and flexibility. The interactive adjustment capability further allows geoscientists to align model outcomes with geological expertise, producing more reliable and geologically meaningful facies classifications.