Abstract
This paper focuses on integrating physical observations and techniques into analytical models to improve well performance predictions in the Midland Basin. By incorporating physical and geological insights through feature engineering and hybrid modeling approaches, we aimed to enhance the accuracy and generalizability of various machine learning models. The study explored three key areas: the development of a physics-informed multivariate regression (MVR) model, the integration of a hybrid, fully-coupled reservoir and fracture simulator with machine learning, and the incorporation of dynamic production and pressure variables into a time-series multivariate regression (DMVR) model. The results demonstrate that combining physical principles with data-driven models can lead to improved understanding for well spacing, completion strategies, and reservoir management decisions.
history matching, geologist, complex reservoir, modeling & simulation, data mining, reservoir simulation, flow in porous media, fracturing materials, geology, proppant
Hydraulic Fracturing, Well & Reservoir Surveillance and Monitoring, Reservoir Characterization, Reservoir Fluid Dynamics, Reservoir Simulation, Formation Evaluation & Management, Unconventional and Complex Reservoirs, Information Management and Systems, Fracturing materials (fluids, proppant), Flow in porous media