3D Structural Geomechanics and Machine Learning Based Petrophysical Evaluation of Proven and Mature Oilfield in Upper Indus Basin, Pakistan

ABSTRACT: Machine learning (ML) has revolutionized petrophysical analysis by providing advanced tools to efficiently interpret well-log data and predict reservoir properties. In this study, key well logs, including Gamma Ray (GR), Resistivity (LLD, LLS, and MSFL), Neutron Porosity (NPHI), Bulk Density (RHOB), and Sonic (DT), were utilized to evaluate reservoir characteristics in the Meyal Oilfield. Supervised ML algorithms, such as Random Forest Regressor (RFR), Extra Trees Regressor (ETR), and Support Vector Machines (SVM), were deployed to predict critical properties, including porosity, permeability, water saturation, and shale volume. Dimensionality reduction via Principal Component Analysis (PCA) and clustering techniques like K-Means further enhanced feature selection and geological interpretation. The application of ensemble learning and artificial neural networks (ANNs) demonstrated exceptional accuracy in automating well-log interpretation, surpassing traditional methods in efficiency and precision. In addition, seismic data analysis was conducted using 2D seismic lines, integrating ML-predicted petrophysical properties with structural interpretation. Horizons corresponding to the Sakesar and Chorgali formations were delineated, revealing structural traps and fault systems crucial for hydrocarbon accumulation. This study underscores the transformative role of ML in subsurface reservoir characterization, highlighting its potential to optimize hydrocarbon exploration and production strategies in complex geological environments like the Potwar Basin. 1. INTRODUCTION Artificial neural networks (NNs) have been used in a study to forecast how well CO2 foam flooding will work to improve oil recovery on a lab scale. Using petrophysical data, created a model that uses artificial intelligence (AI) to forecast the porosity and permeability of petroleum reservoirs in addition to reservoir characteristics modeling, many scientists have created data-driven methods for wax deposition prediction and applied advanced machine learning techniques to the problems encountered in engineering, construction, and other industries (Akkurt et al., 2018). A comparative analysis of the integration of different machine-learning methods used to estimate the energy efficiency of buildings’ heating loads for smart city design. ML-based models to forecast permeability impairment due to scale deposition were studied, along with a comparison of various ML techniques for estimating the permeability and porosity of oil reservoirs using petrophysical logs (Al-Khalifa et al., 2020). When compared to traditional methods, it was found that the Extra Trees Regressor performed exceptionally well in estimating the volume of shale and porosities, while RFR and DTC were the most effective in modeling Sw and facies. This is ascribed to its capacity to accurately detect patterns in the training data and hence model reservoir features (Zhang et al., 2021). With a time-efficient approach and optimized results, suggested machine learning algorithms have effectively addressed the drawbacks of traditional methods, including generalization and data range, for petrophysical prediction without requiring extensive use of geological or lithological characteristics of the reservoir formation as shown in Figure 1 By adding more realistic data (core) and comprehensive data sets from all over the world with difficult reservoirs, this strategy can be further enhanced (Najwa et al., 2023) With a variety of inputs, machine learning algorithms can gain experience without being specifically designed to do so. Classification, continuous value prediction, and performance or event forecasting are among the predictions that may be made using the built model (Bader et al., 2019).