A comprehensive simulation study tested and analysed the sensitivity of dynamic connectivity in turbidite channel reservoirs to a large number of stratigraphic and engineering parameters. The study showed that subseismic shale architecture has a significant effect on reservoir connectivity. However, representing the complete spectrum of fine-scale architectural details in full-field simulation models is beyond the limits of existing computational capabilities. Previous work demonstrated that incorporating geologically based pseudo-relative permeabilities into relatively coarse full-field reservoir models renders practically intractable simulation cases tractable. We developed a methodology for generating pseudo-relative permeabilities at multiple geological scales, incorporating the effect of channel architecture and reservoir connectivity into fast simulation models.
We describe a dynamic modelling workflow that integrates geologically based pseudo-relative permeabilities into a two-stage automatic history-matching algorithm. The history-matching problem is posed as one of data conditioning in the Bayesian framework. We show the application of the workflow to a channelized turbidite reservoir in West Africa. It is demonstrated that multiple geologically consistent models that are conditioned to production data can be generated rapidly thanks to optimally coarse simulation models that capture the effect of subseismic channel architecture on recovery behaviour, and run efficiently as the forward model within a Bayesian inference framework. Proof-of-concept tests carried out using field data indicate that the history-matched models predict well-by-well future recovery response with good accuracy.