3D Source Rock Characterization, Energy Storage Estimation and Machine-Learning Algorithms for CO2 Sequestration in Depleted Oil Reservoirs for Enhanced Oil Recovery: An Integrated Approach for CO2-EOR

ABSTRACT: The integration of advanced computational techniques with traditional reservoir engineering practices has revolutionized the field of enhanced oil recovery (EOR), particularly in the context of carbon sequestration. This study integrates machine learning (ML) algorithms and 3D visualization techniques to optimize CO2 sequestration and enhanced oil recovery (CO2-EOR) in the depleted Balkasar Oil Field, located in Pakistan's Potwar Basin. The application of ML, including neural networks and decision trees, allowed for the detailed analysis of extensive datasets comprising well logs, seismic surveys, and petrophysical measurements. The algorithms uncovered complex relationships between key reservoir properties, such as porosity, permeability, and hydrocarbon saturation, enabling precise predictions of hydrocarbon-bearing zones and dynamic fluid behaviors. A 3D geomechanical grid was constructed to visualize subsurface heterogeneities, including fault zones and fluid distribution, aiding in the identification of optimal CO2 injection zones. Simulations conducted on this grid revealed a recovery efficiency enhancement exceeding 30%, confirming the potential for significant hydrocarbon recovery. Moreover, pressure monitoring and CO2 plume migration predictions validated the integrity of sequestration, with stable storage observed under varying injection scenarios. The study's findings underscore the transformative potential of integrating advanced computational tools with traditional reservoir engineering methods. For instance, dynamic modeling in the Chorgali and Sakesar formations identified zones with high hydrocarbon retention, ensuring strategic injection planning. The multivariate fitting method demonstrated robust predictions of organic carbon content, essential for assessing the reservoir's remaining hydrocarbon potential. 1. INTRODUCTION The transition toward sustainable energy practices necessitates innovative strategies that simultaneously address the imperatives of carbon reduction and efficient hydrocarbon recovery. The global energy sector is facing increasing pressure to mitigate climate change while meeting growing energy demands (Earlougher, 1977). This process involves injecting CO2 into depleted oil fields, where it reduces oil viscosity, improving its flow to production wells (Bourdet, 2002). ML models, such as neural networks and decision trees, can uncover hidden patterns in the data, improving the accuracy of reservoir predictions and optimizing CO2-EOR strategies (Manrique et al., 2007).