AI Fast-Tracking Infill Well Identification Under Uncertainty for a Complex Deep Gas Field and Accelerating Field Development Decision

Abstract Objective/Scope R’ field is a tight, deep gas-condensate volatile-oil reservoir, in a geologically complex volcanic depositional setting. Vertical and deviated wells penetrate several zones, multi-stage fractured making it commercially viable. Ramp-up and sustaining production at the planned plateau production rate through intensive development drilling campaigns was a challenge. The primarily transient production profiles and recoveries per well are highly variable and the identification of sweet spots and good quality reservoirs(pay) is fraught with uncertainty. There was an urgency to fast-track optimized drilling schedules that maximize return on investment while sustaining production. This called for both a better quantification of risks and a higher speed in subsurface modeling and field development planning. Methods, Procedures, Process An integrated reservoir modeling workflow with probabilistic dynamic model calibration approach was designed for uncertainty quantification and sweet-spot identification. The static and dynamic models maintained a high resolution to retain fidelity and allow precision in modeling the hydraulic fractures, and phase behavior effects. The geological and completion uncertainties number exceeded 1500, however fast simulation algorithms on high performance computing was leveraged to process multiple realizations. A high-fidelity simulator capable of modeling time-variant, bi-wing multistage fractures at a field scale, Machine Learning (ML) proxies of the dynamic simulation models were the innovative technologies utilized to make the uncertainty workflow practical and affordable. The ML proxies were used to screen and rank uncertain parameters and generate posterior probability distributions used to sample representative models for planning purposes. Probabilistic maps of connectivity were generated using streamline simulation to guide both history-matching and well-placement activities. Decision criteria were developed that integrated model-based technical and economic risk attributes with other analytical attributes to help screen and rank well locations with the high potential and lowest risk. Results, Observations, Conclusions Novel/Additive Information An innovative approach of ML-assisted history matching and probabilistic forecasting without compromising the model fidelity proved invaluable. Collaborative and agile approach of rapid prototyping and power of cloud computing enabled evaluation of 1000s of field development scenarios and thereby fast-tracking the decision-making process.