Pattern Recognition and Data-Driven Analytics for Fast and Accurate Replication of Complex Numerical Reservoir Models at the Grid Block Level

Abstract

Reservoir simulation models are used extensively to model complex physics associated with fluid flow in porous media. Such models are usually large with high computational cost. The size and computational footprint of these models make it impractical to perform comprehensive studies which involve thousands of simulation runs. Uncertainty analysis associated with the geological model and field development planning are good examples of such studies. In order to address this problem, efforts have been made to develop proxy models which can be used as a substitute for a complex reservoir simulation model in order to reproduce the outputs of the reservoir models in short periods of time (seconds). In this study, by using artificial intelligence techniques a Grid-Based Surrogate Reservoir Model (SRM G) is developed. Gridbased SRM is a replica of the complex reservoir simulation models that is trained, calibrated and validated to accurately reproduce grid block level results. This technology is applied to a CO 2 sequestration project in Australia. This paper presents the development of the reservoir simulation model and the Grid-based SRM. The SRM is able to generate pressure and gas saturation at the grid block level. The results demonstrate that this technique is capable of generating the reservoir simulation output very accurately within seconds.

A data-driven smart proxy model for a comprehensive reservoir simulation

“…The proxy models results cannot mimic the numerical simulation results with 100% accuracy, but within the amount of time required to run these models, the outputs generated have a very acceptable range of error. Reducing the computational time from hours and sometimes days to seconds makes these models significantly competent and attractive to the reservoir engineers [29].…”

Assisted History Matching Using Pattern Recognition Technology

“…Since the advent of SRMs in 2006 ( Mohaghegh, et al) many successful examples of their applications have been published Jalali, et al, 2009;Amini, et al, 2012;Amini, et al, 2014;Shahkarami, et al, 2014a). Mohaghegh et al (2012a;2012b) have discussed the results of several projects involving surrogate reservoir models for the fast track analysis of numerical simulation models.…”

“…Mohaghegh et al (2012a;2012b) have discussed the results of several projects involving surrogate reservoir models for the fast track analysis of numerical simulation models. Other publications regarding the SRMs can be found in variety of reference materials (Mohaghegh, 2009;Mohaghegh, 2011;Mohaghegh, 2014;Shahkarami, et al, 2014a;Amini, et al, 2014).…”

Smart Proxy Modeling of SACROC CO2-EOR

“…Mohaghegh describes SRM as an "ensemble of multiple, interconnected neuro-fuzzy systems that are trained to adaptively learn the fluid flow behavior from a multi-well, multilayer reservoir simulation model, such that they can reproduce results similar to those of the reservoir simulation model (with high accuracy) in real-time" [15]. Since 2006, SRM as a rapid replica of a numerical simulation model with quite high accuracy has been applied and validated in different case studies [16][17][18][19][20][21][22]. SRM can be categorized in well-based [17][18][19]21,23] or grid-based types [16,20,24] depending on the objective or the output of the model.…”

“…Since 2006, SRM as a rapid replica of a numerical simulation model with quite high accuracy has been applied and validated in different case studies [16][17][18][19][20][21][22]. SRM can be categorized in well-based [17][18][19]21,23] or grid-based types [16,20,24] depending on the objective or the output of the model. In a well-based SRM, the objective is to mimic the reservoir response at the well location in terms of production (or injection).…”