Petrophysical Characterization of Rock Properties Through Pressure Clustering: A Machine Learning Approach

Abstract This study demonstrates the effectiveness of machine learning (ML) in characterizing reservoir rock properties using pressure clusters—zones with similar pressure regimes and flow dynamics—in clastic sandstone reservoirs of the Niger Delta. Traditional methods for delineating hydrocarbon flow units (HFUs), such as core sampling and lab-based techniques, are time-intensive and costly. Here, we propose an ML-driven approach that bypasses these limitations by leveraging pressure clusters to map spatial variability in reservoir properties. A dataset integrating petrophysical well logs and pressure measurements (Bottom-Hole Pressure/Temperature tests) was analyzed using Random Forest (RF) and Gaussian Mixture Models (GMM). These algorithms classified petrophysical properties with 99% accuracy and identified two primary HFUs. To validate robustness, results were cross-checked against conventional Reservoir Quality Index (RQI) and Flow Zone Indicator (FZI) methods (Al-Ajmi and Holditch, 2000). Both approaches consistently identified the same HFUs, with RQI-FZI log-log plots showing strong correlation (R2 = 96%). This synergy between ML and traditional techniques highlights ML's potential to accelerate reservoir characterization while maintaining scientific rigor. By rapidly identifying homogeneous flow/storage zones—even in unlogged intervals—ML reduces reliance on extensive coring, cutting costs and time. The success of pattern recognition in heterogeneous reservoirs underscores ML's role as a transformative tool that complements—rather than replaces—established methods. This work bridges innovation and reliability, offering a scalable framework for strategic reservoir management.