AI-Driven Robust Decision-Making Framework for Optimized Oil and Gas Reservoir Development

Optimal petroleum reservoir development decisions require balancing conflicting objectives, for instance, maximizing economic returns, minimizing environmental impact, and optimizing production efficiency, while considering geological and economic uncertainties. This study introduces an integrated framework that combines machine learning (ML) with multi-objective optimization (MOO) to enhance decision-making and value creation, improve solution robustness, and increase transparency in complex and uncertain reservoir management contexts. The proposed framework employs the non-dominated sorting genetic algorithm II (NSGA-II) to generate an approximate Pareto front under uncertainty. A machine learning-assisted approach is incorporated to refine non-dominated alternative selection, which uses a simplex-based weight generation technique to systematically explore trade-offs among multiple competing objectives. Additionally, decision tree visualization simplifies complex decision structures, aiding interpretability without compromising solution accuracy. The methodology is applied to the UNISIM-II-D benchmark, a Brazilian pre-salt reservoir model, where six objectives are analyzed across four production strategy alternatives. The NSGA-II framework identifies four distinct production strategies (named as strategies A, B, C and D or Neco Max) under geological and economic uncertainty, all of which are non-dominated, posing a challenge for decision-makers. A systematic weighting mechanism is applied to rank the alternatives based on trade-offs, leading to the selection of Strategy Neco Max as the most favorable option. The ML component effectively translates high-dimensional weight spaces into interpretable decision rules, facilitating informed, high-quality decision-making. The integrated approach improves decision robustness, reduces cognitive complexity, and enhances consensus-building in capital-intensive reservoir projects. These findings demonstrate the framework's potential to optimize production strategies while including uncertainty in petroleum reservoir management. This study advances decision analysis in petroleum reservoir management by integrating data-driven methodologies, including NSGA-II and machine learning-assisted decision support, with a systematic weighting approach to create a novel, scalable framework for managing multi-objective trade-offs in complex industrial applications. Applied to the UNISIM-II-D benchmark, this methodology optimizes production strategies like Strategy Nevco Max, offering flexibility in uncertain environments and contributing to energy resource optimization. The decision-quality chain justifies selecting Nevco Max by leveraging comprehensive knowledge and creative alternatives, framing it around economic, environmental, and efficiency goals, ensuring clear values and sound reasoning through decision tree visualizations, and demonstrating commitment to actionable, robust outcomes.