A Novel High-Speed AI-Physics Hybrid Model to Predict the Reservoir Behavior; A Case Study

Abstract

Abstract Reservoir simulation is widely applied to model and manage subsurface flow operations. However, due to the nonlinear nature of the governing equations and the multiscale character of the geological description, executing numerical simulations can be expensive and increasingly slow. In this study, we propose an innovative high-speed simulation method to build a model that is capable to predict the field's behavior. In this study, a novel AI-Physics hybrid model was proposed for reservoir simulation. Classical numerical simulation and deep learning neural networks are coupled together to create an AI-Physics hybrid reservoir simulation. As a result, a model was obtained with predictive capabilities to forecast the field's behavior. Then, we combined AI-Physics history training with blind test prediction calculation of remaining oil maps. Finally, forecast scenario definitions based on the remaining oil map were created by the AI-Physic model. The proposed high-speed simulation model can reduce the history matching and scenario assessment time by 90 to 95%. According to its capabilities, three improved forecast scenarios were created based on a predefined scenario. These improved scenarios can produce a significant million standard barrels more oil than the original development scenario within three years. This technology eliminates limitations for multiple scenario assessments. In our AI hybrid model, the power of dynamic reservoir simulation is combined with a modern machine learning approach to "Evergreen" forecasts in reservoir assets. Consequently, the simulation resulted in a sub-optimal shortcut between model updates and inconsistencies in production forecasting. Moreover, applying deep learning methods to focus on the critical reservoir properties intelligently leads to tremendous time-saving in the static model update life cycle. In fact, with this novel simulation that we implemented, the new production data could be incorporated within minutes to regenerate more reliable and up-to-date forecasts. This simulation generates ‘up-to-date remaining hydrocarbon maps interactively, so the operator can continuously optimize the infill drilling locations between Field Development Plan cycles.