The clastic Field B offshore Malaysia has been under production since 1971 and a new full field simulation study was undertaken for a comparative evaluation of late life waterflooding versus redevelopment with only infill wells.
The area of interest spans 5 neighboring fault blocks, each consisting of up to 52 stacked clastic reservoirs, and more than 250 unique fluid contact sets (OWC & GOC). In addition, facies and petrophysical property distribution, communication between fault blocks, saturation end-point variations, and aquifer strength are significant uncertainties. Allocated production is also a major challenge due to commingled production in ~60 mainly dual string wells with up to 10 commingled sands per string.
Historically, simulation studies were approached deterministically, by anchoring static inputs to a single static geological realization and initializing the simulation model with uncertainty parameters at base values. During history matching the model would be gradually and subjectively conditioned at various levels to best reproduce the observed production data. The result of this time-consuming trial and error process is invariably a single non-unique conditioned model, largely anchored to the original static model, with manual, subjective and biased modifications to the static and dynamic input data.
To address the challenges associated with the large number of uncertainty parameters and to obtain a range of better history-matched models faster, an ensemble of 100 equiprobable reservoir models was built in an integrated workflow incorporating multi-disciplinary inputs to holistically represent all reservoir uncertainties and associated ranges. The integrated ensemble-based modeling project was undertaken in a cloud-based cognitive Exploration and Production (E&P) digital platform, an efficient digital ecosystem able to cater for computationally intensive parallel run of ensemble cases to evaluate the impact of all combinations of uncertainties.
The models were conditioned to historical production, pressure and saturation data using an iterative ensemble-based data assimilation algorithm capable of handling large number of reservoir uncertainty parameters simultaneously in a consistent manner, on both a local level (e.g. facies and petrophysical properties) and a global/regional level (e.g. rock curves, PVT model, fault transmissibilities & threshold pressure, contacts & aquifer size), which reduces modelers’ bias and ensures a high level of static geological & dynamic data integrity. While assimilating production data and calibrating the ensemble, the initial uncertainty range of all the parameters implemented in the study also reduce naturally.
After 4 iterations, an ensemble of acceptable quality was obtained and by using the historical cumulative oil production as a criterion, the 15 best conditioned realizations were selected for forecasting. The forecast scenario of interest can be evaluated considering the remaining static and dynamic uncertainties through the sub-samples of selected models.
The resource volume range of the existing producing wells estimated using Decline Curve Analysis (DCA) is well aligned with the range from the forecasts of the history matched (conditioned) models and is thus a good assurance of model quality. The Expected Ultimate Recovery (EUR) range from infill wells as predicted by the conditioned models is also aligned with recent drilling campaign results.
The ensemble-based approach has proven to be an efficient way to condition geologically-plausible, equiprobable realizations to the historical data in just 7 months – a period significantly shorter than that usually required to manually history-match a single deterministic realization.
The ensemble-based methodology has enabled a timely decision to prefer infill drilling over late-field life water flooding.
