Identification and Localization of Remaining Mobile Oil in Mature Fields Using Integrated Physics-Based and AI/ML Workflow: A Case Study

Abstract

Abstract Identification and localization of remaining mobile oil are fundamental components of mature-field reservoir management. This task becomes increasingly challenging when large portfolios contain numerous fields, reservoirs, fault blocks, and producing intervals with varying levels of data completeness and subsurface understanding. Conventional field-by-field studies based on detailed static and dynamic reservoir modelling can provide reliable results but require substantial time, data preparation, and technical resources. Consequently, a rapid and consistent screening methodology is required to identify the fields, reservoirs, and well locations that should be prioritized for more detailed subsurface maturation. This paper presents an integrated physics-based and artificial intelligence/machine learning workflow developed for portfolio-scale screening and localization of remaining mobile oil. The methodology follows a sequential two-step approach in which analytical reservoir-engineering methods are applied first to rank fields and reservoirs, followed by production-data-driven AI/ML analysis to identify underperforming wells and localize potential opportunity areas. The automated workflow consists of two main stages: (1) Analytical models: zone/block level Buckley Leverett and recovery factor benchmark screening: each field zone / block production, reservoir, and fluid data are consolidated, quality controlled, and corrected for missing or non-physical values using regional correlations, field averages, and analogue. Remaining mobile oil is then estimated using two independent analytical approaches. The first applies Buckley–Leverett fractional-flow calculations to estimate movable oil after water displacement. The second compares the planned recovery factor, derived from production and booked resources, with a benchmark recovery factor calculated from displacement and sweep-efficiency relationships. The combined results are adjusted for existing production and booked volumes, filtered using a minimum opportunity-volume threshold, and used to rank fields and reservoir zones. (2) AI/ML models: producer-level remaining oil localization: Production histories from the highest-ranked fields are utilized to calculate > 50 well performance indicators such as early-time production, cumulative oil, decline behaviour, water cut, gas-oil ratio, production duration and uptime, etc. Multiple regression-based machine-learning models are trained to predict expected ultimate recovery from well performance indicators. Underperforming wells are identified from the positive difference between ML-predicted and actual well EUR. The calculated gap between predicted by ML and actual well EUR, remaining-production potential, is normalized to the historical performance distribution of infill wells in the field drilled for last years. Each well estimated potential is integrated in bubble maps to evaluate today high water cut, high GOR, and low pressure risks to define and rank remaining-mobile-oil opportunity areas. The methodology was applied to a large mature-field portfolio comprising > 150 fields, > 5000 field–zone–block evaluation areas, and > 3000 well-completion intervals. The analytical screening identified nine fields with the highest additional 2C potential for detailed well-level evaluation. The AI/ML stage subsequently identified high-potential underperforming producers, grouped the well-level potential into reservoir-scale opportunity areas, and supported the identification of "low hanging fruits" (well-reopening, well reactivation) and infill-drilling candidates. Opportunities were further ranked with today high water cut, high GOR, ow reservoir pressure risks, minimum economic-volume thresholds, and wells accessibility. The workflow was validated through comparison with calibrated, history-matched dynamic reservoir models and ongoing field studies remaining mobile oil interpretations. Validation included comparison of the remaining mobile oil location in identified opportunity areas and ranking of the most potential reservoir zones and blocks today. Thus the paper introduces a scalable workflow that combines classical reservoir engineering analytics with AI/ML well-performance screening to accelerate mature-field 2C opportunity maturation and shorten the identification to maturation to execution cycle.