New Era of History Matching and Probabilistic Forecasting—A Case Study

Abstract

This paper presents a case study using optimization technology to improve the reliability of reservoir simulation models. Global optimization techniques have been applied to assist the history matching (HM) performed. Evolutionary Algorithms and deterministic optimization schemes are integrated into a workflow controlling a large number of parallel reservoir simulations. Results are analyzed and compared to traditional HM to identify the potential added value and increased efficiency. Focus is given to the reservoir engineer who is in full control over the optimization process and interacts frequently as he/she acquires more information and improved understanding of subsurface uncertainties. The case study demonstrates the benefit and added value of utilizing a computer assisted HM workflow. Pitfalls of using a traditional workflow are illustrated. The paper focuses on a gas condensate reservoir where the most important well had proven to be difficult to match using a traditional manual HM approach. The sensitivities of a large number of input parameters were investigated, and an optimization scheme was set up in order to generate alternative HM realizations of the simulation model. Alternative models were generated using a wide variety of input parameter combinations. The entire study was conducted over a two week period, where approximately 7000 full field reservoir simulations were run. A large number of simulation models were matched within the uncertainty of the observed data, and these acceptable matches gave a wide range of recoverable reserves from the field. The first match that was found proved to be outside of the initial uncertainty range, clearly verifying the need for more than one HM model to cover the uncertainty found in most producing reservoirs. This illustrates the non-uniqueness of simulation models and the danger of relying on one deterministic model. Introduction During the last few years the reservoir engineers have fallen dramatically behind the rest of the sub-surface community. While significant resources have been spent on developing efficient tools and workflows for generating alternative geological realizations, the production forecasting workflow has not improved accordingly. As the work performed by the reservoir engineer most often is limited by the time frame available, the traditional manual workflow used is simply too slow to utilize the information gathered by the other disciplines. Relief is fortunately on the way as a new era is rising within the discipline of reservoir engineering. The improvements are driven by the new computational power available for simulation studies. The hardware cost has been dramatically reduced the last few years, and will continue to do so. LINUX clusters provide the opportunity to run parallel and distributed reservoir simulations at affordable prices. A new generation of reservoir simulators with improved capacity for parallel processing and improved solvers is also being brought to the market. Furthermore, recent advances in multi-core processors will change the way engineers work with computers. Multi-core processoring is the future of computing1–3, which will enable engineers to run a large number of simulations on their desktop PC or laptop. The evolution of multi-core processoring is a technical and commercial challenge for simulation software vendors regarding licensing issues and pricing policy. These technological advances provide new opportunities for use of computer assisted history matching. While the automatic HM tools developed during the 1990's and earlier where employing simple gradient based optimization methods and running the reservoir simulations sequentially, more robust optimization methods are now available. They allow the engineer to launch simulations in parallel and using global optimization techniques4–10.

Bridging the Gap Between Deterministic and Probabilistic Uncertainty Quantification Using Advanced Proxy Based Methods

“…Although they may generate an ensemble, that ensemble is biased and does not characterize a probabilistic distribution, even if some weighting scheme is used based on a likelihood (Gross, Honarkhah, & Chen, 2011) (Holmes, McVay, & Senel, 2007). (Erbas & Christie, 2007a) (Erbas & Christie, 2007b) (Selberg, Ludvigsen, Harneshaug, & Diab, 2006) (Subbe, Christie, & Sambridge, 2003) (Hamman, Buettner, & Caldwell, 2003) (Schaaf, Coureaud, & Labat, 2009) (Lach, et al, 2005) (Hajizadeh, 2011). We generally refer to this kind of approach as 'possibilistic'.…”

Assisted History Matching for Petroleum Reservoirs in the Social Computing Era

“…In the recent past, studies have proven that the use of assisted methods together with the optimization theory can considerably reduce the time needed to calibrate a model (Cullik et al, 2006;Selberg et al, 2006;Fokker et al, 2013). Furthermore, assisted methods can provide multiple possible solutions, which offer a much more representative evaluation of the uncertainty associated with the production forecasts.…”

“…Sensitivity-based methods, such as Gauss-Newton and the Least Square method (LSQR) have also been tested. The gradientbased method is intrinsically sequential and cannot exploit efficiently parallel architectures; on the other hand, the global optimization methods,in general,are easily parallelizable and can greatly benefit from Science Publications AJAS distributed architectures, which allow running several simulations simultaneously (Selberg et al, 2006). Current developments in multi-core processors allow parallelization of numerical codes and, as a consequence, speed up of the calculations (Oliveira et al, 2013).…”

Benefits and Limitations of Assisted History Matching

“…Furthermore, researchers have shown great interest for the Ensamble Kalman filter (EnKF), a data assimilation technique often used in other scientific fields like ocean prediction systems (Evensen 2009). The gradient-based method is intrinsically sequential and cannot exploit efficiently parallel architectures; on the other hand, in general the global optimization methods are easily parallelizable and can greatly benefit from distributed architectures that can run several simulations at the same time (Selberg et al 2006). However, software for reservoir simulation does not seem to currently offer good flexibility in terms of licensing for distributed computing.…”