Although upscaling has been extensively investigated, a quantitative relationship between the multi-million block geological model and the upscaled flow model is likely to be lost during the history match process, primarily due to parameter modification on an upscaled model. In this paper, we propose a new history match procedure combined with an innovative faster upscaling technique. In the proposed procedure, after each history run, parameter modification for the next run was done on the detailed geological model in close collaboration with geologist. This loop was made possible with a new faster upscaling technique, distinct against the existing method, by adopting empirical correlations of computing upscaled absolute and relative permeabilities. A capillary-limit upscaling technique was applied for the case study, since a target reservoir is capillary dominant. In the case study, history match was conducted on a sector of an oil-wet carbonate reservoir under waterflooding for 20 years. In each modification, the revised geological model of three million blocks was upscaled to a flow model of 60 thousand blocks in a process independent manner with a scale-up factor of 50. The CPU speed-up factor was approximately 200: 10 days for the detailed model with massive parallel computing and an hour for the upscaled model with a personal computer. Even with such significant CPU time reduction, it was confirmed that the upscaled model still maintained the simulation quality of the detailed geological model. Introduction Recently, the size of a geological model has been becoming huge, e.g., multi-million-gridblock model, in order to describe the heterogeneity as much as possible. Direct transfer of such a huge geological model (detailed geological model) to a flow model would result in a too long CPU time in a flow simulation and be impossible even for nowadays computer capacity in terms of the required memory. In general, the objective of upscaling is to reduce the number of gridblocks of a flow model so that a flow simulation can be conducted in a reasonable range of a computation time. The requirement for upscaling is that the important characteristics of heterogeneity and multiphase flow are to be retained in the resultant upscaled flow model made up of the fewer number of gridblocks than the detailed geological model. In other words, upscaling is one of such techniques that assign representative properties to each simulation block of a coarse grid model given that a finely gridded model is available. In the literature, there are two major upscaling approaches: dynamic approaches and effective properties approaches (Ekrann and Dale 1992). Dynamic approaches adopt pseudo- functions based on the simulation results of an entire reservoir model or representative portion of a fine grid model. Thus, these approaches give accurate results limited to an imposed flow condition from which pseudofunctions are derived. However, the derivation is too time-consuming to upscale very large geological models. In addition, the pseudofunctions have many other deficiencies as criticized by Barker and Thiebeau (1997). On the other hand, effective properties approaches average flow properties from the information on sub-grid scale heterogeneity alone in which the local flow regime assumed to be extreme: viscous, gravity, or capillary dominated. These approaches, which work on the isolated coarse grid block, enable us to compute effective properties much faster than dynamic approaches and give accurate results if the appropriate assumption is made for the actual flow field.

Equations of state (EOS) are perhaps the most abused and mis-used concept in compositional reservoir simulation. It is believed that the EOS are often need adjusting or tuning and the well-tuned EOS can capture all phase behavior in the reservoir simulation. However, the well-tuned EOS can not calculate appropriately in the reservoir simulation because the matching objectives, the existing PVT experiments, are not enough to cover all necessary pressure, temperature, and components. An EOS for CO2, hydrocarbon gas and sour gas injection is tuned for a reservoir fluid offshore Abu Dhabi. The universal EOS for gas injection, EOS-U8, was originally developed as EOS-H18 for hydrocarbon gas injection (both sour and sweet). After the EOS-H18 was developed, additional PVT experiments, swelling tests with CO2 and other two hydrocarbon gases, were conducted and the EOS-U8 was tuned in order to capture the phase behavior of the reservoir fluid and all candidate injection gases. The new, tuned EOS is available for evaluating gas injection, including direct comparison between injection gases. Our tuning procedure for an EOS, including laboratory data selection, is applicable to other reservoir fluids. The Flash calculation is compared utilizing the tuned EOS using all or selected laboratory data. Limited laboratory data may mislead the tuning process of the EOS parameters for extensive application of the EOS. The methods that gave the best match are also discussed. Three commercial software for generating EOS parameters were compared for performance, accuracy, and ease of use. The advantages and disadvantages of three commercial EOS PVT programs are compared. The interface and function of the programs are also evaluated. Three EOS based PVT programs show different results with high concentration of CO2 or H2S using the tuned EOS parameters. Therefore, special cares are required for importing EOS parameters form other EOS programs. The most appropriate ways to use the EOS developed by current software products are discussed. The results from this comparison study of EOS based PVT programs are useful for selecting PVT programs. Finally three EOS (EOS-H8, C8, U8) were compared in the compositional reservoir simulation. The tuned EOS with the selected laboratory data shows different results in the reservoir simulation using the EOS-U8. It is recommended that EOS should be tuned utilizing as much as PVT experiments. Introduction Equations of State (EOS) are one of the most important and sensitive factors in reservoir simulation for gas injection study. In 1998, intensive phase behavior study was conducted for a limestone reservoir in a large field of Abu Dhabi offshore. Special PVT tests were conducted using two hydrocarbon gases and one sour gas. The first EOS, EOSH18, was developed, utilizing Software-A, in order to capture the phase behavior of the reservoir fluid with three different components of injection gases. In the next year, additional phase behavior study was conducted for the reservoir fluid with sour gas, hydrocarbon gas and CO2. The EOS-H18 was updated as EOS-U8 in order to capture the phase behavior of the reservoir fluid with all six gases. In this updating process, it was found that the original EOS might not be compatible with Software-B. The EOS-U8 was developed with the Software-B, which was selected because of the compatibility with the Compositional Simulation Software-D. In this study, the matching results of EOS using all or selected data are discussed. Then, comparison results of three commercial software for generating EOS parameters are presented. Finally, influence of misused EOS is discussed in the point of the compositional reservoir simulation.