Abstract:
This study addresses the crucial challenge of characterizing thin gas sand reservoirs in the Lower Ranikot/Khadro Formation of Pakistan's Lower Indus Basin, a reservoir with varying thicknesses 4 to 7 m below seismic resolution. Previous studies have struggled to produce precise results due to reservoir heterogeneity, data limitations, and associated uncertainties. An optimized, integrated approach, combining seismic attributes, petrophysical properties, advanced machine learning (ML) algorithms, and continuous wavelet transform (CWT) addresses thin gas sand facies and pore pressure challenges comprehensively. Among several employed ML algorithms gradient boosting regressor (GBR) accurately predicted thin sands (>90%), reducing uncertainty in hydrocarbon-bearing sand distribution. A delicate ML approach has been broadly applied to analyze the potential and robustly interpret well-logs while addressing the associated challenges. Support vector machine (One-class-SVM) helps to reduce outliers with great certainty while the missing log's sonic and density are precisely predicted via GBR and extra tree regressor (ETR) with the highest R2 respectively. Likewise, random forest regressor (RFR) performed exceptionally well for water saturation modeling expressing the highest 0.93 correlation among ML and conventional results. Finally, the decision tree classifier (DTC) modeled reservoir facies with the best 91% accuracy and 0.935 F1 measures at the blind well. Additionally, an optimized workflow generates high-frequency acoustic impedance synthetics by utilizing a deep neural network (DNN) integrated with CWT components at the reservoir level vis-a-vis validating the results with existing geological facies to resolve thin beds without introducing noise. The shale layers of the formation are quite problematic and complex geological variations exhibit pore pressure discrepancy making drilling operations crucial. Among all conventional methods for pore pressure prediction, GBR integrated with CWT has provided very good results after validation. The study characterizes reservoirs below seismic resolution, enabling more efficient resource exploration and development. It outperforms previously done conventional approaches by delivering higher accuracy, reducing uncertainty, and unlocking valuable insights using advanced ML and CWT techniques. It offers broad applicability to other complex, thin-bed reservoirs worldwide, optimizing field development and maximizing hydrocarbon recovery.
