Initialising Dynamic Reservoir Models for History Matching Using Pre-Production 3D Seismic Data

Abstract

Abstract Stochastic reservoir modeling is a common practice in the energy industry, and is widely used for hydrocarbon reserves estimation, targeting new producer/injector locations, and production profile forecasting with flow simulators. Due to its high spatial coverage, 3D seismic data plays a critical role for defining the reservoir geometry, and for constraining physical property modeling. However, integration of 3D and time-lapse 4D seismic data into the reservoir model history matching process poses a significant challenge due to the frequent mismatch between the initial reservoir model, the reservoir geology, and the pre-production (baseline) seismic. Therefore, a key step in a reservoir performance study is the preconditioning of the initial reservoir model to equally honor both the geological knowledge and the baseline seismic data. In this study, we investigate issues that have a significant impact on the (mis)match of the initial reservoir model with the geological and geophysical data. Specifically, we address the following questions:Which of the common methods to stochastic litho-facies modelling produce reservoir models that best match the baseline 3D seismic data after seismic modelling?How are the results affected by the presence of noise in the observable data, and by the low vertical resolution of seismic data compared to logs?What is the effect of geostatistical variogram parameters on the results?How do these methods perform on object-oriented reservoir models? The results of this study indicate that a method based on the probability of litho-facies distribution given by P-wave impedance in a stochastic modeling process yields the best match to the reference model, even in the presence of noise in the dataset. The effect of variogram parameters on the seismically-constrained litho-facies modeling process is also demonstrated.