Perpustakaan Digital ITB

Water saturation calculation in reservoir rocks is crucial for estimating hydrocarbon reserves and optimizing production strategies. This thesis investigates the prediction of rock types and water saturation using advanced methods, focusing on uncored intervals. The study integrates the Pore Geometry Structure (PGS) method with Artificial Neural Networks (ANN) to predict rock types where core data is unavailable. The Lambda method, combined with curve fitting techniques, is employed to derive the Saturation Height Function (SHF) from capillary pressure data. The research begins with a comprehensive review of existing methodologies and theories related to rock typing, saturation height functions, and machine learning applications in petrophysics. Rock types are initially classified using the Pore Geometry Structure (PGS) method and Artificial Neutral Network (ANN) models are trained to predict rock types in uncored intervals based on log data. As a result Artificial Neural Networks achieved an accuracy of 91% on training data and 88% on testing data, indicating that the model is effective for rock typing with some minor prediction errors. Subsequently, the Lambda method is applied to fit capillary pressure curves obtained from laboratory data. The resulting SHF is compared against log-derived water saturation to assess accuracy and performance. Curve fitting techniques are employed to optimize the Lambda function parameters, demonstrating improved prediction capabilities over traditional methods. The measurement of water saturation using the lambda function resulted in an error of approximately 27% compared to log-water saturation. In contrast, the lambda method with a curve fitting approach produced an error of 11% compared to log-water saturation, indicating that curve fitting can reduce the error in water saturation calculations.