Automated Well and Reservoir Management Using Hybrid Physics and Data-Driven Models - Case Study

Production optimization of hydrocarbon fields is a complex and challenging task that requires accurate reservoir knowledge and efficient data management. Traditional surveillance methods for well and reservoir performance are manual, interpretive and do not scale over several wells with high data volumes to make timely decisions. Can hybrid models that blend physics and data-driven methods be used for automatic performance tracking in relevant time, early detection of anomalous events, actionable opportunity identification, and continuous system optimization? A digital solution encompassing various reservoir and production workflows was implemented to improve asset operational efficiency. Event detection algorithms for well shut-ins and stable flow periods are used to catalog planned and unplanned events based on routine field operations. Shut-in analysis is performed automatically to extract key well parameters (e.g., productivity index, average reservoir pressure) and reservoir diagnostics using a hybrid model (physics-informed data-driven approach). Any abnormal deviations in well performance such as productivity decline, liquid loading, choke erosion, and water or gas breakthrough are flagged as anomalies that enable exception-based surveillance. Well, inflows are updated based on changes in operating points using steady-state detection algorithms. Subsequently, the integrated production system comprising of wells and pipeline network is calibrated continuously and used for production optimization and short-term business forecasting by evaluating several scenarios. In this paper, we discuss the application of the integrated hybrid modeling system to shallow water assets. The system allowed to quickly evaluate and identify several value opportunities for the fields with significant reservoir uncertainties. Event detection algorithms increased model calibration data density available for shut-in analysis by over 200%. An unsupervised learning method for choke erosion prediction was validated through hindcasting that could have avoided expensive choke washouts by early intervention. Self-calibrating models enabled continuous and improved scenario optimization for integrated asset management with 10-15% uplift in field production. Model quality improved as the frequency of model calibration increased from manual updates twice a year to once a week with the automated system. A reduced physics-based reservoir model was used to estimate connected reservoir volumes and inter-well connectivity with reasonable confidence for active reservoir surveillance as an alternative to traditional reservoir simulation. The field application demonstrates the effectiveness of the integrated hybrid modeling system for automatic performance tracking, early detection of anomalous events, actionable opportunity identification, and continuous system optimization. It helps to improve asset operational efficiency, optimize production, and reduce costs by avoiding expensive choke washouts through early intervention. The solution utilizes a new generation of hybrid modeling tools that enables continuous asset optimization. Furthermore, the solution augments the capability of engineers by providing a practical way of combining data-driven methods with our understanding of physics.