Digital Twin for Real-time Reservoir Optimization under Uncertainty

Thesis Abstract

The study addresses the critical challenge of achieving real-time, uncertainty-aware reservoir optimization in volatile oil and gas environments through an integrated digital twin framework. Traditional reservoir management relies on static models and retrospective data, which limit responsiveness to dynamic reservoir behavior and multi-source uncertainty, including geological heterogeneity, fluid properties, and operational disturbances. The aim is to develop and validate a scalable digital twin that fuses high-fidelity reservoir simulations with real-time data streams to continuously calibrate predictive models and optimize field-wide decisions under uncertainty. Specific objectives include (1) designing a modular digital twin architecture that integrates seismic, production, and completion data with probabilistic calibration; (2) implementing real-time data assimilation using ensemble Kalman filters and particle filters to update reservoir states and uncertain parameters; (3) embedding stochastic optimization and robust control strategies within the twin to maximize expected Net Present Value (NPV) under multiple scenarios; (4) evaluating the impact of model-plant integration on operational decisions such as artificial lift, waterflood management, and well placement; and (5) conducting a comprehensive sensitivity and risk analysis to delineate critical drivers of performance. The methodological approach adopts a mixed-methods design anchored in computational experimentation and case-study validation. The population comprises synthetic and field-scale reservoir models representative of mature carbonate and sandstone reservoirs with heterogeneous permeability fields. A representative sample set of 12 field-scale cases and 4 synthetic benchmarks will be used to test scalability and transferability. Data collection leverages multi-physics simulations (CMG, ECLIPSE) for baseline reservoir behavior, coupled with real-time telemetry streams from actual production facilities and IoT sensors. Instrumentation includes electronic well logs, downhole pressure and temperature measurements, surface facility data, and seismic-derived porosity/permeability estimates. Model calibration employs ensemble-based data assimilation (ETKF and HBEF variants) to update reservoir states and uncertain parameters (porosity, permeability, relative permeability curves) at hourly timesteps. The analytical framework uses stochastic optimization, specifically robust model predictive control (RMPC) and Bayesian optimization, to derive policy decisions that maximize expected NPV while constraining risk via conditional value-at-risk (CVaR). Uncertainty propagation is analyzed through polynomial chaos expansion (PCE) and scenario-based Monte Carlo simulations. The study also implements a digital governance layer to ensure data integrity, security, and traceability aligned with ISO 27001 standards. Analytical techniques include regression analysis to relate data assimilation improvements to production performance, ANOVA to compare performance across scenarios, and time-series analysis to evaluate twin responsiveness. Model validation employs cross-validation against historical field campaigns and out-of-sample forecasting tests, with performance metrics including prediction RMSE, NPV uplift, and constraint violation rates. The expected findings indicate that the digital twin reduces reservoir state estimation error by 25–40%, accelerates decision cycles by 30–50%, and yields a 6–12% uplift in cumulative oil production under uncertainty across case studies. The twin’s ability to re-calibrate in near-real-time is anticipated to improve the accuracy of water cut forecasts and infill-well optimization, thereby mitigating excessive pumping costs and suboptimal infill locations. The study contributes to knowledge by operationalizing a scalable, uncertainty-aware digital twin for reservoir optimization, bridging gap between advanced data assimilation, stochastic optimization, and field deployment. It advances theory in digital twin governance for complex subsurface systems, demonstrates applicability of RMPC and Bayesian optimization in petroleum decision-making, and provides a replicable framework for integrating real-time telemetry with legacy reservoir models. The main conclusion is that a well-architected digital twin substantially enhances resilience and economic performance of reservoir operations under geological and operational uncertainty. Recommendations include adopting standardized data interfaces for cross-operator interoperability, extending the twin to include chemical EOR analytics, and pursuing iterative deployment strategies across stages of field development to maximize long-term value.