Japan National Oil Corporation (JNOC) is conducting a survey on underground gas storage (UGS) using inert gas as cushion gas, which makes good use of assets of depleted gas or gas condensate reservoirs. Since the Japanese regulation for a heating value of city gas is extremely strict, a heating value and hence composition of gas withdrawn from those reservoirs must be carefully predicted to properly design UGS projects. JNOC has developed an UGS simulator, which can rigorously evaluate a heating value of withdrawn gas taking into consideration all the major phenomena that possibly affect composition of withdrawn gas. A conventional cubic equation of state (EOS) based 3-D compositional simulatorand a PVT simulator were improved to construct the UGS simulator equipped withthe functions for calculating physical phenomena:molecular diffusion andvelocity dependent dispersion,three-phase flash including dissolution of gaseous components into water and vaporization of water,adsorption of gaseous components onto a rock surface andturbulence and Klinkenberg effects. In addition, new functions of MPFA (Multi-Point Flux Approximation)and TVD (Total Variation Diminishing) scheme with local grid refinement in acorner point geometry were incorporated into the simulator to rigorously calculate a fluid flux with permeability tensors and to reduce numerical dispersion, respectively. Furthermore, functions to calculate temperature distributions both in a reservoir and wellbore were added to examine the effectof cooling caused by gas injection and adiabatic expansion on fluid flow and gas composition. Each function newly developed for the UGS simulator was then validated through test runs using laboratory test data. Field scale simulations for hypothetical reservoirs were also conducted to confirm the simulator's performance as wellas to examine the effect of cushion gas volume, working gas volume, reservoir heterogeneity and in situ gas composition on withdrawn gas composition. This paper describes the development and validation of the UGS simulator followed by the results of field scale simulation runs. Introduction Since natural gas was first stored underground in 1915, an increasing number of fields have been utilized for UGS for the purpose of peak shaving and stableenergy supply. According to 1991/1992 statistics, there are 550 UGS fields inthe world, 425 of which utilize depleted gas or oil reservoirs [1]. A conventional black oil type simulator or modified black oil type simulator that enables distinction between injection and in situ gas phases has been most commonly used for evaluation/prediction of UGS reservoir performances [2, 3,4]. A black oil type simulator, however, is not satisfactory to rigorously calculate composition of withdrawn gas and hence a heating value of it. Major phenomena that possibly affect composition of withdrawn gas are:*mixing between in situ fluids and injection gas,*diffusion/dispersion of cushion gas when inert gas is used as cushiongas,*dissolution of gaseous components into water phase and water vaporizationinto gas phase,*adsorption/desorption of gaseous components onto a reservoir rocksurface,*turbulent flow caused by high rate withdrawal/injection and Klinkenberg effect,*hysteresis of gas-liquid capillary pressure and relative permeability, and*reservoir temperature change that results from cooling by injecting vaporized LNG and adiabatic expansion of gas. In addition, numerical errors such as numerical dispersion and discretization errors should be minimized for accurately reproducing these physical phenomena.

Higher order numerical schemes are necessary to get good front resolution when modelling reservoirs using practical block sizes. Methods for such schemes presented to date have focussed on situations where the grids are somewhat regular. To be useful, implementations are required that can apply in situations when the gridding is complex, such as when corner point grids are involved that have non-regular cells, grid refinements are used, and when faults exist. This work discusses the implementation of higher order accurate methods in a corner point setting when refinements and faulting are present, and includes a discussion of how to maintain numerical stability. The implementation is carried out in an equation of state-based fully compositional reservoir simulator that uses complex corner point grids with faulting and refinements, thereby making higher order methods available for field scale reservoir modelling. The numerical schemes use two point flux calculations and can be up to second order accurate in space in smooth regions, the latter being regions where the various component, phase saturation and pressure distributions (and hence phase velocities) are smoothly varying. A Total Variation Diminishing (TVD) flux limiter is required to maintain numerical stability. Such limiters are applied to the inter-cell flows in each component's mass balance equation to control throughput and to ensure overall stability of the resulting numerical scheme. The application of flux limiters also guarantees that the computed properties remain within their physical ranges. TVD limiters must be applied carefully when complex grids are encountered as cells have varying dimensions and can be partially contacted by neighbouring grid cells on several faces. Special consideration needs to be given to how the limiters are to be evaluated in these situations. This usage extends TVD technology beyond what has already been done in reservoir simulation, and makes its use practical for complicated field scale models. The results of several simulations are demonstrated. In particular, the ability of the techniques to better resolve sharp fronts can be seen. This leads to a more accurate prediction of overall fluid movements, including fluid contamination and mixing, breakthrough timing, and front and/or fluid bank movement in the reservoir. A more precise evaluation of ultimate recovery for the reservoir is obtained, which leads to the opportunity to improve overall recovery. The techniques described here can greatly enhance the accuracy of compositional reservoir simulation, and it is shown how these capabilities can be brought to the realm of field scale modelling with complex corner point grids. Introduction Complex corner point grids are often encountered in reservoir simulation. Such grids can be severely faulted and the corner point cells near faults, and elsewhere, can be distorted. Also, extensive refinements may be used. A corner point grid cell starts with eight points, and can be described as the volume contained within surfaces that are stretched across the corner points associated with each of its six faces. Face contact (or near-contact) will be specified as the requirement for flow to occur from cell to cell. The contact can be full (the four face corner points are shared by a neighbouring cell), or partial. Once contacts are established and overlap areas are determined, transmissibilities for computing flows can be defined. Corner point cells are often refined and several nested levels of refinement can be present. Details regarding contact determination and transmissibility calculation are given in [7].

No abstract