Integrating Seismic Facies Analysis and Machine Learning for an Accelerated Regional Geologic Interpretation

This study presents a hybrid machine learning workflow integrating Genetic Algorithms (GA) and Self-Organizing Maps (SOM) to improve seismic facies classification in carbonate reservoirs, characterized by geological complexity and heterogeneity. GA automates seismic horizon picking by optimizing waveform similarity, reducing manual input and enhancing accuracy. Seismic attributes, including instantaneous frequency, phase, Sobel filter, and spectral decomposition, are calculated between GA-defined horizons to focus on geologically meaningful zones. These attributes serve as inputs for SOM clustering, revealing depositional environments, lithological variations, and structural features. Validation with core and well log data ensures geological accuracy and enhances subsurface models. The workflow effectively delineated depositional patterns, identified facies transitions, and refined gross depositional environment maps, extending facies boundaries and providing actionable insights for hydrocarbon exploration and reservoir development. Despite its effectiveness, the study revealed limitations in regions with sparse well or core data, underscoring the need for comprehensive ground-truth datasets, suggesting future integration of supervised learning and expanded attribute sets. By bridging the gap between seismic data and geological understanding, this study offers a practical approach to improving exploration strategies and decision-making processes in complex carbonate reservoirs.