Digital twins for offshore reservoir optimization under real-time data constraints

Thesis Abstract

Offshore reservoir management increasingly relies on digitalized, interconnected data streams to enable real-time decision making; however, the effectiveness of digital twins in such environments is constrained by data latency, incompleteness, and model-plant mismatches that degrade predictive accuracy and operational responsiveness. This study aims to develop and validate a robust digital twin framework for offshore reservoir optimization that operates under real-time data constraints, improving adaptive control of production strategies and reservoir stewardship. Specific objectives are (1) to quantify the impact of data constraints on reservoir performance predictions using a hybrid data-modeling approach; (2) to design a resilient digital twin architecture that integrates real-time sensing, edge analytics, and cloud-based optimization under stringent latency requirements; (3) to implement adaptive model updating and uncertainty quantification through Bayesian updating and ensemble Kalman filtering for rapid assimilation of streaming data; (4) to evaluate decision-support performance against baseline static models across multiple simulated and field-representative scenarios; and (5) to develop governance and operational guidelines for field deployment, including data quality thresholds and fail-safe procedures. The methodology adopts a mixed-methods research design combining quantitative simulation experiments with qualitative expert input for model validation. The population comprises offshore fields with heterogenous geology and production systems; a representative synthetic reservoir dataset (n=10 case studies) is generated to capture a range of heterogeneities, complemented by real-world production data from two mature offshore fields provided under data-sharing agreements. Data collection instruments include high-fidelity reservoir simulators (black-oil and compositional models), real-time surface facility data streams (pressure, temperature, surge rates, choke settings), subsea sensor telemetry, and operator logs. Model development employs a hybrid modeling stack physics-based reservoir simulators for baseline behavior, machine-learning surrogates for fast predictive updates, and probabilistic data assimilation modules (particle filters and ensemble Kalman filters) to fuse streaming data with model states. The analytic framework integrates regression analyses to quantify data constraint effects, Bayesian updating to manage uncertainty, and optimization algorithms (black-box and gradient-based) to derive production strategies. Theoretical grounding draws on the Theory of Constraints and the Bayesian decision theory, with an integration of concepts from digital twin architecture design and cyber-physical system resilience. Data analysis proceeds in three phases. First, sensitivity analyses quantify how data latency, gaps, and noise influence forecast accuracy and control performance, using repeated-measures ANOVA and multivariate regression. Second, real-time assimilation experiments assess convergence properties and robustness of the digital twin under varying data quality, evaluating metrics such as root-mean-square error, forecast horizon, and control regret. Third, scenario-based evaluation compares production and economic performance—net present value, oil recovery factor, and CO2 emissions—across the digital twin-enabled strategy versus conventional planning, employing paired t-tests and bootstrap confidence intervals. Qualitative insights from expert interviews (n=12) with production engineers and reservoir geoscientists are analyzed using thematic analysis to identify practical barriers, data governance requirements, and operational workflows. Expected findings indicate that the proposed digital twin framework maintains stable predictive accuracy within 5–10% relative error under moderate data constraints and delivers measurable improvements in production optimization, reducing decision latency by up to 40% and increasing NPV by 6–12% in simulated scenarios. The integration of adaptive surrogate models and Bayesian updating is anticipated to yield more reliable forecasts during data outages, while ensemble-based uncertainty quantification provides principled risk-aware control actions. The study contributes to knowledge by delivering a technically validated blueprint for real-time, constraint-aware digital twins in offshore reservoirs, including an architectural reference model, data governance guidelines, and a decision-support protocol that links sensing fidelity to operational outcomes. The main conclusion is that, with tight integration of edge analytics, probabilistic data assimilation, and adaptive surrogates, digital twins can robustly enable offshore reservoir optimization under real-time data constraints. Recommendations emphasize investment in low-latency data pipelines, standardized data schemas, rigorous data quality monitoring, and collaboration frameworks between operators, service providers, and researchers to facilitate field deployment and continuous improvement.