Innovative Numerical Simulation of Sandstone-Basement Connectivity and Machine Learning-Driven Production Optimization for Composite Reservoirs

Abstract This study focuses on the sandstone-basement composite reservoirs in the Bongor Basin, West Africa, where sandstone directly overlies the basement without a significant intervening stratum, leading to operational interference between reservoirs. Despite this, stratigraphic principles have guided development, and the lack of understanding of these interactions poses a challenge for efficient, long-term production. Addressing this gap, we develop a novel numerical model to characterize dynamic connectivity and apply machine learning for optimizing production strategies, aiming to enhance reservoir management and hydrocarbon recovery. We employ an integrated approach combining single-porosity and dual-porosity discrete fracture models, leveraging the finite element method for numerical simulation to quantify the degree and direction of interference. Sensitivity analysis is conducted to identify key factors influencing reservoir interference. Additionally, a multilayer neural network model is trained to correlate the reservoir parameters with oil production, facilitating the prediction of yields and optimization of production pressures for both sandstone and basement. The validated model, grounded in an actual production data-driven approach, demonstrates little error in well output ratios of sandstone to basement, highlighting significant pressure drops in sandstone due to fluid attraction by the fractured basement. Key factors influencing interference are identified as basement thickness, well spacing, with thickness positively correlating with interference, and well spacing negatively. To expedite the finite element solution, a surrogate neural network model is introduced, enhancing computational efficiency. Optimal production pressures for balanced flow and peak oil output are identified through this model, combined with a non-gradient optimization algorithm. It suggests that increasing sandstone reservoir pressure differential can improve overall development effectiveness by balancing the flow field. This research develops an innovative numerical model that provides deeper insights into the dynamics of sandstone-basement reservoirs. A machine learning-based surrogate model is employed to optimize production strategies, offering a vital framework for operational monitoring and real-time updating of production parameters in the field.