Perpustakaan Digital ITB

This study proposes an AI-driven approach for reservoir quality mapping, integrating Fuzzy Logic and Fuzzy Pattern Recognition (FPR) within the Top-Down Reservoir Modelling (TDRM) framework. The goal is to identify the most effective method for defining reservoir boundaries and representing quality variations over time. Three methods 4x4 uniform grid, fuzzy boundaries, and gradient-based (linear regression) segmentation are compared for accuracy and consistency. Cumulative production data from 70 wells in the Northwest Java Basin (1970-2025) were processed using Fuzzy C-Means clustering into four quality classes: “poor,” “average,” “good,” and “excellent.” Reservoir segmentation was performed using spatial attributes and tested across multiple time intervals, with blind testing to evaluate robustness. Findings show that the uniform grid method cannot capture actual reservoir complexity, while fuzzy boundaries yield dynamic but overly broad, literature-inconsistent results. The gradient-based method, using slope variations in production curves, provides the most accurate and consistent boundaries, aligning with fuzzy clustering outcomes. Observed quality shifts, such as between “poor” and “average,” were effectively represented, supporting better field development decisions. The study concludes that gradient-based segmentation is the most reliable for boundary delineation, while fuzzy clustering effectively classifies reservoir quality in complex, uncertain conditions.