Page 264 - Intelligent Digital Oil And Gas Fields

214 Intelligent Digital Oil and Gas Fields

• Shirangi and Durlofsky (2015) developed and apply a methodology for
closed-loop field development (CLFD) under geological uncertainty.
A multilevel hybrid optimization, combining metaheuristic techniques
PSO with mesh adaptive direct search (MADS), is validated for simul-
taneous and sequential field development with maximizing NPV as
the objective function. They further propose enhancements in the form
of bi-objective optimization with minimizing the risk objective while
maximizing the expected performance.
• Arnold et al. (2016) propose a comprehensive, field full lifetime
workflow for uncertainty propagation and rigorous optimization of
decision-making under uncertainty, applied to assets with naturally frac-
tured reservoirs. They use the design of experiments (DoE) to perform
sensitivity analyses and distance-based multidimensional scaling (MDS)
to identify and dynamically rank (Maucec et al., 2011) the model can-
didates, multipoint statistics (MPS) to generate updates of the discrete
fracture network (DFN) models in the field appraisal phase, the multi-
objective PSO (MOPSO) for history matching and in development, his-
tory matching and Naı ¨ve Bayes (NA-Bayes) for unbiased estimation of
uncertainty for forecasting and reservoir management phases of the field
lifetime. This novel attempt at technology integration results in <5000
reservoir simulation iterations to attain a robust set of decisions for a
complete field lifetime.

6.3 ADVANCED MODEL CALIBRATION WITH ASSISTED
HISTORY MATCHING

The calibration of reservoir simulation models to dynamic field pro-
duction data, commonly referred to as dynamic data integration or history
matching, is perceived as one of the most time-consuming engineering pro-
cesses in reservoir validation. In a New Technology Magazine article (Cope,
2011), Maucec stated, “History matching must be considered as a bridge
between the reservoir modeling and reservoir simulation.” Traditionally,
reservoir models were manually reconciled with production data, using
good engineering judgment and following workflows based on many years
of experience.
The main disadvantage of the manual history-matching process is that
it disengages the reservoir simulation model from the geological model
and, many times, fails in adequate quantification of reservoir uncertainty.
As a result, manual history matching frequently leads to unrealistic,