Application of Deep Neural Network (DNN) in Reservoir Simulation for Hydraulic Fracturing and Production of Unconventional Wells

Abstract This study focuses on developing a Deep Neural Network (DNN)-based reservoir simulator aimed at improving the efficiency and precision of hydraulic fracturing and production forecasting. Given the complex and nonlinear nature of reservoir behavior, our goal is to implement an automated, scalable solution to effectively navigate these challenges. The proposed DNN simulator integrates diverse data sources, enhancing a comprehensive understanding of reservoir dynamics and supporting real-time decision-making and operational optimization. Reservoir simulation, a key numerical method for predicting fluid behavior in porous media like oil and gas reservoirs, is vital for crafting and refining production strategies in the oil and gas industry. A swift and accurate simulation is essential for effective field development. The study particularly addresses unconventional wells—horizontal wells fractured to facilitate oil and gas flow—where challenges persist in areas like fracture geometry characterization, fluid dynamics between the matrix and fractures, and pressure variation prediction during field development. The scarcity and diversity of data, encompassing both the volume and variety from sources like geology, geophysics, lab tests, and engineering, add complexity to simulating these wells. We explore the use of Deep Neural Networks, a machine learning algorithm capable of deciphering intricate data relationships, in this context. Recent advancements in machine learning have opened avenues for employing DNNs in reservoir simulation, which is a supplement to traditional time-intensive mathematical algorithms. We constructed and simulated both three-dimensional conceptual models and unconventional field models using this approach. The results validate the deep learning method's accuracy and speed, showcasing its supplementary to conventional numerical reservoir simulators.