Simultaneous optimization of multiple objective functions for reservoir modeling

Abstract

We have developed a new method, a synthetic validity test, and a field data demonstration for constructing reservoir models that simultaneously match available seismic, borehole, and geologic data in an optimal way. We combine geostatistical simulation with simultaneous nonlinear stochastic optimization of multiple objective functions, and we do not require the selection of weighting schemes or multiple optimization computational runs. Because each geologic and geophysical data set has its own strengths and weaknesses, realistic models are best obtained by simultaneously matching all of the multiple data constraints. We define a set of objective functions for each of the multiple data sets that are used to constrain the reservoir-modeling optimization. In the examples, we specify an [Formula: see text]-norm misfit to match inverted seismic attribute volumes, an [Formula: see text]-norm to match lithofacies and porosity logs, and the Hausdorff metric to match image patterns in geologic map-based information. We use a geostatistical technique to initialize a population (ensemble) of starting models and run a nonlinear evolutionary algorithm to drive the ensembles of model solutions toward the Pareto optimal front, along which represents the best-compromise set of model solutions that simultaneously satisfy all of the multiple objectives. We tested our method on a 3D synthetic object-oriented reservoir model for which variogram-based simulation techniques typically fail to reproduce realistic models. We applied our method on a producing reservoir located offshore Western Australia. Our results indicate that the new method is capable of producing optimal models of reservoir properties and flow-unit connectivity that are consistent with known reservoir information, including dynamic production data such as pressure interference tests, and 4D seismic monitoring data.

A Review of Geophysical Modeling Based on Particle Swarm Optimization

“…For the sake of completeness, another MOEA adopted in geophysics is the Nondominated Sorting GA (NSGA-III) (Deb and Jain 2014). NSGA has actually been more explored than MOPSO and applied to the inversion of: Raleigh-wave dispersion curves and reflection travel times (Dal Moro and Pipan 2007 ), surface wave dispersion and horizontal-to-vertical spectral ratio (Dal Moro 2010 ), seismic and well-log data for reservoir modeling (Emami Niri and Lumley 2015 ), magnetic resonance and VES data (Akca et al 2014 ), AMT and broad-band MT data (Schnaidt et al 2018 ), and receiver functions, surface wave dispersion and MT data (Moorkamp et al 2011 ).…”