SPE 116113 Analytical Upgridding Method to Preserve Dynamic Flow Behavior

Seyyed Abolfazl Hosseini; Mohan Kelkar, SPE, The University of Tulsa

Copyright 2008, Society of Petroleum Engineers This paper was prepared for presentation at the 2008 SPE Annual Technical Conference and Exhibition held in Denver, Colorado, USA, 2124 September 2008. This paper was selected for presentation by an SPE program committee following review of information contained in an abstract submitted by the author(s). Contents of the paper have not been reviewed by the Society of Petroleum Engineers and are subject to correction by the author(s). The material does not necessarily reflect any position of the Society of Petroleum Engineers, its officers, or members. Electronic reproduction, distribution, or storage of any part of this paper without the written consent of the Society of Petroleum Engineers is prohibited. Permission to reproduce in print is restricted to an abstract of not more than 300 words; illustrations may not be copied. The abstract must contain conspicuous acknowledgment of SPE copyright.

Abstract Geo-cellular model contains millions of grid blocks and needs to be up-scaled before the model can be used as an input for flow simulation. Available techniques for upgridding vary from simple methods such as proportional fractioning to more complicated methods such as maintaining heterogeneities through variance calculation. All these methods are independent of the flow process for which simulation is going to be used, and independent of well configuration. We propose a new upgridding method which is based on the pressure profile for each two individual pair of grid blocks in the fine scale model. It is well established that more complex the flow process, more the detailed level of heterogeneity is needed in simulation model. In general, ideal upscaling is the process which preserves the pressure profile from the fine scale model under the applicable flow process. In our method we try to upgridd the geological model using the some basic ideas of flow in porous media to make the model applicable for flow simulations. However for getting a better match with respect to fine scale model one has to use proper property averaging techniques. We used the new method for 2D and 3D case and both for single phase and two phase flow cases. The method is currently developed for single phase flow; however, results in both cases defer from that of methods using variation of a property as coarsening criterion. The procedure is static and hence very efficient but at the same time includes the dynamic features in it. The grid blocks (or layers) are combined in a way so that the difference between fine scale and coarse scale pressure profile is minimized. Also this new method provides two new criterions that let us to choose the optimum number of the layers more accurately. Also these criterions can provide an insight about the level of heterogeneity of the model and effectiveness of the layering design. We compare the results of our method with variation based methods and proportional method and show that, for the same number of layers, the proposed method better captures the results of the fine scale model compared to King et als method. We show that the layer merging not only depends on the variation in the permeability, but also on relative magnitude of the permeability. We also show that new method can incorporate some dynamic variables like grid block thickness and size. Introduction The geological models have very complex structures and because of the high resolution in reservoir characterization accordingly these models could have millions of grid blocks which is not efficient to run for flow simulations. Upscaling is a technique that reduces a geological model to a simulation model i.e. reduces the number of grid blocks to the level that for a reasonable amount of time and memory computers can handle it. With the available computers we can only afford to run the flow simulation for the models of size 105 grid blocks while the common geological models dramatically exceed this number of grid blocks [1]. On the other hand if one intended to do the auto history matching for the model, the size has to be reduced significantly to make the history matching iterations feasible even if new history matching techniques like EnKF method have been used. In most the cases because of the high resolution in vertical direction we have detailed layering while our areal resolution is not as fine as vertical resolution. This results to the fact that vertical upgridding is an important issue in the process of upscaling. Some difficulties regarding the areal upscaling have been addressed by King et.al [2].The process of up scaling can be divided to different steps followed by each other: upgridding and averaging the new properties for coarse grid blocks. There have been several papers regarding the both upgridding [2-5] and calculating the new properties [6-8]. This paper engaged with the problem of the upgridding and derivations are based on single phase flow assumptions though there are papers addressing the issues related to the two phase flow upscaling [1, 9]. This paper intends to answer two questions in this context. First; in which sequence one has to combine the layers? Second; where we should stop the process of the coarsening? i.e. what is optimum number of the layers for a geological model that can regenerate the dynamics of the fine

SPE 116113

scale model? The earliest method to do such an upgridding was capacity fraction technique which ranks the permeabilities in order of magnitude regardless of their location [10]. Following those early works Testerman [4] used a statistical technique to identify and describe the naturally accruing zones in a reservoir. Furthermore he used those data to correlate these zones from well to well. In their early work they introduced the idea of the within grid blocks variation and between grid blocks variation which has been used by others to develop new techniques. Stern and Dawson [5] introduced a new coarsening technique. They combine the layers to have the smallest change in geological model properties and they use some calculations of the sweep efficiency for unit mobility displacements to find the optimum number of the layers but in their algorithm still one need to do some iterations and check at each step of coarsening for sweep efficiencies. Generating such a coarse grid models remains one of the active subjects of the research. Most of the works done to upgridd the geological models are static based methods and there are two main categories in these methods: Sequential coarsening [23, 5] and sequential refinement [4]. it worth to mention that to find the optimum layering design there has been lots of paper published but in most of the papers there is no clear answer to the question that what is the optimum number of the layers and in cases that there is, no real field simulation results has been provided to confirm the correctness of the methods. On the other hand in most of the methods user has to provide a threshold to be able to see the number of the optimum layers and it makes the process to be trial and error and empirical. Fincham et. al. [11] and meddaugh [12] have been implemented some of the current upscaling techniques for field cases and discussed their limitation and impact on dynamic response of the models. In this paper, we have tried to provide such a measure that can help to find the number of layers more accurately and provide some extra information to the user about the effectiveness of the layering design with respect to the case of having no design i.e. proportional upgridding. By design we mean fractions of the fine scale model that have to be combined in fine scale model to reach to the coarse model. Also in variation based methods user can choose different properties as an input though there are several papers suggesting the best representative variables but practically the first choice of the users is to use permeability. On the other hand, calculations in most of the upgridding methods are statical and based on the variation (or variance) of a property. In new method due to the use of the Darcy equation we integrate more parameters like grid block size and thickness. In this paper, by giving some simple numerical examples we will try to demonstrate some of obvious advantages of the new method. We will not address the problem of the property averaging and we will focus on the upgridding problem, even though we believe that the problem of the finding new properties is an important task has to be done after upgridding and in some cases it has more effect than choosing the optimum design. There are some recent papers discussing such techniques both on averaging the permeability [8] or more recently averaging the transmissibility [6]. It worth to mention that from our experience we believe that for a geological model if model A has the best possible layering design and model B is any optimum layering design, as far as the layering design technique has properly chosen, dynamic response of model B will be very close to one of the model A and as far as we can get model B with fast algorithms, probably it doesnt worth to use prolonged methods to get closer to model A. So after finding the proper coarsened model if one needs a better match with respect to fine scale model, the answer lays on the finding the best averaging technique for new grids, mainly finding the proper transmissibilities. In this paper we discuss a new approach based on the Darcy equation and we derive a criterion for combination of the layers. For our application we are defining a new set of boundary conditions and we make some assumptions to simplify the final equations. The details of the work and algorithm will be discussed later in this paper. Algorithm is fast and fairly simple and there is no need to run the fine scale simulation. The scale of the calculations is (NX*NY*NZ) and comparable to the Kings method and King has been provided a comparison about the scale of the calculations for different methods [2]. This new method like Kings method can also be used to generate unstructured grid blocks although running the flow simulation for unstructured models is very expensive except if a reliable high performance unstructured simulator is available. Our methods defers from the old methods that use the variation for their work by utilizing the Darcys equation as its criterion. We are defining a new total error for any two layer combination. This total error term is a multiplication of different error terms. For this paper we only introduce the errors introduced to total error coming from pressure profile difference and physical difference of the layers. This new method has the ability to be modified to account for more options like cross flow effect which is under the investigation. The order of material which is going to be presented in this paper is as follow: first we describe our approach and assumptions to find the new criterion for coarsening process. Second we will discuss two new concepts of design factor and error per layer that by making proper graphs of these variable we can obtain an idea about the optimum number of the layers and the extend that layering design is going to be effective in whole process of the coarsening. At the end we will provide some simple numerical examples to demonstrate the main differences with old methods and finally application of this new method will be presented for a real field case. For real field cases we have taken the proper measures to deal with the undefined and non-porous grid blocks. In most of the cases we will compare our results with proportional and variation based methods.

SPE 116113

Theory and Definitions This method is using the same idea of the static methods to combine the layers to keep the advantage of being fast but at the same time by using the Darcy equation and pressure profile of the adjacent grid blocks includes some dynamic effects to the model. The very first advantage with respect to the old methods is that it combines the layers not only by looking at the difference in the variation of a property like permeability but also one will see that the relative difference between the variables is important. There are some papers suggesting other variable except than permeability to be used in variation based methods like local velocities because velocity directly relates to permeability you can see that this problem still is there. For a model with dimensions of Nx*Ny*Nz, we define a total error term which is the multiplication of different error terms. It has the flexibility to have more terms in it as one finds the other factors affecting the process of upgridding e.g. including the effect of cross flow, adverse mobility ratios, and so on.
E
k
T

i, j

E t 1 E t 2 E t 3 ...

(1)

Where i =1, 2, 3, Nx; j =1, 2, 3, Ny and k =1, 2, 3, Nz-1 Assume that we have two adjacent grid blocks that has to be combined. Subscript of 1 stands for grid block (i, j, k) and subscript 2 stands for grid block (i, j, k+1). We assume that we have single phase flow according to the Darcy steady state equation along with the following boundary conditions for these two grid blocks. For grid block (i, j, k): @ X = 0 ; P = P0 and @ X=L ; P=P1 For grid block (i, j, k+1): @ X = 0 ; P = P0 and @ X=L ; P=P2 According to the Darcy equation we will have for each of these grid blocks:
P1 = P0 Cte Q1 ( X X 0 ) H 1 Dy K 1

(2)

P2 = P0

Cte Q 2 ( X X 0 ) H 2 Dy K 2

(3)

Now if we replace these two grid blocks with a single coarsened grid block with following boundary conditions: For grid block (I, J, K): @ X = 0 ; P = P0 and @ X=L ; P=Pu Then the new pressure profile in the coarsened model will be:
Pu = P0 Cte Qt ( X X 0 ) H t Dy K u

(4)

We separately treat each two grid blocks with the same boundary conditions because we dont want to solve the full system of the equations for pressures. On the other hand we dont want to make this process to be dependent on the well locations. Now we are defining the following error term introduced due the coarsening:
E1 = ( P u ) dX 1 P
0 L

(5) (6)

E2 = ( P2 Pu )dX
0

And we define the first error term in equation (1) as follow:

2 Et1 = E12 + E2

(7)

Now by assuming the arithmetic average for the permeability in coarsened model and assuming that:
Q1 = H1 dy Q2 H 2 dy

(8)

i.e. having the same constant velocities in both grids we can rewrite equation 6 as:
Et1 =
2 2 ( K1 K 2 ) 2 ( K12 H12 + K 2 H2 ) Q 2 2 Cte 2 ( ) ( ) dx 4 2 2 K1 K 2 ( K1 H1 + K 2 H 2 ) 2 H 2 dy 2

(9)

We already assumed that velocity is constant and if the sizes of grid blocks (i, j, k) and (i, j, k+1) in x and y directions are the same, then equation (8) can be rewritten as follow:
E t1 =
2 2 ( K 1 K 2 ) 2 ( K 12 H 12 + K 2 H2 ) 2 2 K1 K 2 ( K1 H 1 + K 2 H 2 ) 2

SPE 116113

(10) In equation (10) we assumed that the length of the grid blocks are the same which is the case for the most of the fine scale models but in the case of having the fine model which contains grid blocks with different lengths on can keep the length term in equation (10) also we know that the assumption of same velocities is not true in some of the cases but we made this assumption to remove the dependence of the final equation to flow rates so it makes the final equation to be treated as a constant number for each two adjacent grid block. Now we define the second error term knowing that the velocity for a grid block is proportional to its permeability so as first choice we define the following error term.
Et 2 = (V1 V2 ) 2 ( K1 K 2 ) 2

(11)

But one can chose different parameters here to represent the velocities like ones used by King et. al. This error term not only to some extend relieves the assumption of the equal velocities but also weights the physical differences between two grid blocks in the final equation. Having equations (10) and (11) and using i, j, k indexing, we can rewrite equation (1) in following format:
( K (i, j , k ) 2 H (i, j , k ) 2 + K (i, j , k + 1) 2 H (i, j , k + 1) 2 ) ( K (i, j, k ) K (i, j, k + 1))4 k ET = 2 ( K (i, j, k ) K (i, j , k + 1))2 i, j ( K (i, j , k ) H (i, j , k ) + K (i, j , k + 1) H (i, j , k + 1))

(12)

Here you can see some obvious advantages of the method. We have the length and thickness of the grid blocks in the equation that you cannot see in variation based methods. Later on by giving some simple numerical examples we will try to show the differences that may arise from these features. In a map based sense we calculate the error for each two grid block from adjacent layers and sum over the X, Y plan. So ET1 reflects the error will be introduced to the model if we combine the layer 1 and 2 and so on. The rest of the algorithm is fairly simple. We just keep all the calculated error values and combine the layers with minimum error. For the next steps of the coarsening we just revisit the error values in adjacent layers to the combined layer, by simply adding the errors. This algorithm is very fast hence effective. This model can be easily generalized for the case that Ky is different than Kx by just simply adding the same equation except replacing K by Ky in equation (12). Choosing the optimum number of the layers In addition to the above criterion for coarsening the geological model we define a new variable that reflects the effectiveness of the layering design with respect to the case that there is no design i. e. proportional coarsening. We will introduce new parameters that will help us to find a reasonable margin for the required number of the layers in the coarse model and also at the same time obtain some extra information about the level of the heterogeneity of the model that we are dealing with. To find such parameters we tried to use the available error values and design fractions and make the process independent of doing any kind of extra calculations e.g. running stream line simulations or sweep efficiency calculations. We define the best design (BD) as the case that we make the biggest contrast in the model while coarsening which means that coarsening starts from top layer and keeps coarsening from top to bottom. Notice that BD is just an imaginary case and does not exist in reality. We also define no design (ND) case as proportional coarsening which means at each step we equally divide the fine scale layers to find the coarse model. Also we have our out put as optimum design (OD) that provides the optimum fractions that has to be combined. Now we define deign factor (DF) as follow:
DFi =

f f
i i

OD i BD i

f i ND f i ND

(13)

Design factor is changing between 0 and 1. The closer the value of the DF to one is the closer optimum design to best design. The DF value gives information about the quality of the layering design and reservoir. At the same time if this variable has higher values it means that our coarsening process is effective and that reservoir has distinct layers that can be separated. Also if DF values as we coarsen the model decreases sharply it may suggest that we are dealing with a more heterogeneous reservoir but in the case of the fine scale model with distinct layers one will see a gradual decrease of the DF value. We also provide another parameter to find the optimum number of the layers. For this purpose at each step of the coarsening we calculate the error per number of the combined layers at that particular step. Then we look at the change of these values and when ever a big change in this value happens we choose that point as a potential point for optimum number of the layers to stop the coarsening process. This means if the value of the error per layer sharply increases at a particular step then we have added a large amount of the error to that particular coarse grid i. e. we have homogenized a major heterogeneity into a coarse grid. One can expect to see lots of the picks in real field cases. Practically we are looking to coarsen the fine scale model as much as possible. Our final suggestion is to choose the pick points which have a DF value as close as possible to

SPE 116113

0.5 and bigger than 0.4. We believe that for the DF values bigger than 0.5 you can upgrid the model without losing the major dynamic responses of the fine scale model. if DF values are smaller than 0.3 one can not expect to gain a lot just by using the design but we still suggest to use the design in those cases. Another piece of information that one can obtain from this graph is that as your fine scale model is coarsening, if you can see lots of picks prior to the coarse model that you have chose it means you are dealing with a severer heterogeneous reservoir. We have examined this criterion for very simple cases and it gives the reasonable results. Data and Results We will present some simple numerical cases to clear the differences of the new method with variation based methods. We have used Kings method to compare our result. In Kings method local velocities according to their proposed method has been calculated and used. As we mentioned earlier one of the first differences of the method is that it combines the layers not only by looking at their differences but also incorporates some more items of the flow system. Consider the case that we have two vertically neighbor grid blocks. Assume that we have same rock type and same fractional flow curves for two adjacent grid blocks and almost same porosities. Both grid blocks with same thickness of 1 ft and K1=100 md. In Fig. 1 you can see how the error changes with respect to the different values of the K2 for different methods. Practically for layer one it means, it is better to be combined with K2=150 than K2=55 even if difference is bigger in the first case.
K1=100 md; H1=1 ft; H2=1 ft

200

1.4

Error from difference

160

1.2

Error from equation (10)

Error from difference

1 Error from equation (10)

120

0.8

80

0.6

0.4 40 0.2

0 20 40 60 80 100 120 140 160 K2 md 180 200 220 240 260 280

0 300

Fig. 1 Comparison of the error equations for new method and difference based method.

Our first example shows that only using the difference of the parameters can cause to find a layering design which is not the best possible design. Assume a simple 5 layer model with constant permeability for each layer. Assume permeability values from top to bottom are K1=102, K2=120, K3=141, K4=120 and K5=101 md. In this case thicknesses of the all layers are equal. We want to upgrid this model to three layers using the Kings method and new method. Model has size of 20*20* 5 with four injectors at the corners and one producer at the center. All the injectors are under constant bottom hole pressure of 4000 psi and the producer is under constant bottom hole pressure constraint of 2100 psi. Based on the Kings method the best design has fractions of 2, 1, 2 i.e. to find the coarse model we combine the top two and bottom two layers and keep the middle layer untouched. Results from the new method suggest different fractions as 1, 3, 1. In Fig. 2 you can see the fine scale and coarse scale models. This example provided to emphasize the very first differences between the methods that arise in very simple cases. We have used the commercial ECLIPS software to do the flow simulations.

Fig. 2 Fine scale and coarse models design. From left to right; Fine, Coarse King and Coarse new models.

SPE 116113

Fig. 3 compares the results from Kings method with new method. As one may notice even if that we have derived our equation based on a simple steady state Darcy equation it can match filed for three phase. We have water flooding in the reservoir along with the fact that we allow the pressure to fall below bubble point pressure and produce some gas. You can see even in this case we can capture the break trough time better than the other layering design.

Fig.3 Comparison of the simulation results for fine scale and coarse scale models.

Another advantage of the new method is that it can incorporate for the fact that maybe grid blocks in finescale model do not have same thickness. Next example illustrates how this can change the results and by using the variation based methods you can not capture this effect. In this example again assume same reservoir but assume that the fine scale has 3 layers and we want to upgrid the model to two layers. Assume permeability values from top to bottom are K1=152, K2=100 and K3=150 md. But thicknesses in this case are not equal and H1= 1, H2= 1 and H3= 3 ft, we want to upgrid this model to two layers using the Kings method and new method. Model has size of 20*20* 3. Based on the kings method the best design has fractions of 1, 2 i.e. to find the coarse model we combine bottom two layers and keep the top layer untouched. Results from the new method suggest different fractions as 2, 1. In Fig. 4 you can see the fine scale and coarse scale models. In Kings method even if the thickness of the third layer is 10 times more than the first layer it will combine it with second layer as far as they have closer permeability values ( assuming the same rock type).

Fig. 4 The effect of thickness. From left to right; Fine, Coarse King and Coarse new models.

SPE 116113

Fig. 5 compares the results from Kings method with new method. We have better match for both cumulative gas and oil productions.

Fig.5 Effect of the thickness on upgridding.

Now with another simple example we demonstrate how design factor and change of the error per layer can be used to choose the optimum number of the layers. First a simple model will be provided and then a complete analysis on the real filed case will be presented.
Optimum layer selection 1 Design factor/Change of error per layer 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 1 6 Number of the layers 11
Design Factor Change of error per layer

Fig.6 Graph of design factor and change error per layer. In this case pick is at five layers and DF also is around 0.5.

SPE 116113

Consider a fourteen layers model with constant permabilities for each layer and equal thicknesses for all layers. Assume permeability values from top to bottom are K1=20, K2=10, K3=25, K4=1, K5=1200, K6=1000, K7=25, K8=20, K9=30, K10=40, K11=30, K12=45, K13=1 and K14=2 md. We have chosen these values intentionally because we can guess that optimum number of the layers in this case is about 5 layers. If we use these data as an input for the upgridding program we will have the Fig. 6 as an output plus the design fractions for each step of coarsening. We have tried other values for permeabilities and it gives a reasonable answer for each case. Error per layer has been normalized to make the graph easier to understand. As you may notice we dont have lots of picks before the optimum layer and design factor is above 0.4. As we discussed earlier it means that using the design will help to have better result with respect to the proportional design and this model had a distinct layer design and wasnt a very heterogeneous case. We can use this case to elaborate how we find the DF value at each step. For this particular example a coarse model with optimum design has fractions of 3, 1, 2, 6, 2 while no design case has fractions of 3, 3, 3, 3, 2 and best design has fractions of 10, 1, 1, 1, 1. Now we present the implementation of our method for a real field case. The size of the model is 33*33*575 and it is a clastic turbidite type reservoir with three distinct zones. We have used a five spot pattern water flooding scenario with constant bottom hole constraints for both injectors and producers. In Fig. 7 you can see a schematic of the field. For the final graphs some normalizations and filtrations has been done to make the graph readable. Fig.8 is the out put of the model for design factor and change of the error per layer. As we mentioned earlier we will see more picks in the real field cases due to the fact that real reservoirs are more heterogeneous than the simple synthetic cases. According to the graph we choose 78 layers which are about 13% of the original number of the layers to run the flow simulations. We will compare our results with King and proportional methods. In this special case water viscosity of 0.4 and oil viscosity of 0.6 has been considered and appropriate relative permeability data for Kings method had been used. Also we will provide the results for a model with 40 layers (next pick in the graph with the DF smaller than 0.4) just to prove that if by further coarsening to DF values less than 0.4, we are passing the optimum point or not.

Fig.7 Permeability distribution for the reservoir with 575 layers.

In our simulations we assume the isotropy for permeability in all three directions. Having vertical permeability will cause some cross flow in the reservoir that we believe helps us to coarsen the model even more. This effect is under investigation but all the results provided in this paper assumes Kz=Kx. A cross section of the reservoir in Fig. 9 provided to have a qualitative check of the different methods. In this particular reservoir two high permeability sections are distinct and what Kings method does is that it keeps those layers untouched while it over homogenizes middle and bottom part of the reservoir because of their bad quality. Basically Kings method gives a very high resolution to high permeability layers while over homogenizing other parts of reservoir that can cause some problems. Our method looks that provides a reasonable out put for this reservoir and major heterogeneities has been captured. Fig. 10 is the simulation results for this reservoir. As we would expect from the cross section view of the models Kings method can not capture the fine scale dynamics and even in this case it is worse than proportional case but this is not the case for all of the reservoirs. From the simulation responses you can see that our method can capture the fine scale response closely.

SPE 116113

1 0.9 0.8 DF/ Change of error per layer 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 3 33 63 93 123 153 183 213 243 273 303 333 363 393 423 453 483 513 543 573 Number of layers

DF

Change of error per layer

Fig.8 Design factor and change of error per layer for field case. This graph suggests about 78 layers for this model also because we dont have many of picks prior to the optimum point we would expect a good match with fine scale model.

At the end we compare the coarse models with 78 layers and 40 layers. For both case we use our method to find the optimum design fractions. In Fig. 11 you can see that with 40 layers we have lost the match between the fine scale and the coarse scale model but interestingly it is almost as good as the proportional model with 78 layers which shows how finding the proper design can be effective. Notice that DF value for 40 layers is bigger than 0.3 so we would expect to have better results with respect to the proportional design case but it is smaller than 0.4 that suggest we cant expect to have very good match with respect to the fine scale model.

Fig.9 Qualitative comparison of 78 layers coarse models to 575 layers fine model. Fine scale, proportional, Kings and new method coarse models. Kings method over homogenizes some parts of the reservoir to preserve the high permeability layers.

10

SPE 116113

Fig.10 Comparison of the simulation results for the proportional, King and new method coarse models with 78 layers fine scale model of 575 layers.

Fig.11 Comparison of the simulation results for new method coarse models with 78 and 40 layers with fine scale model of 575 layers.

SPE 116113

11

Conclusions We developed a new analytical criterion to coarsen the geological models. This method is fast hence effective. Using the simple Darcy steady state equation we derived a criterion that can incorporate thickness and size of the grid blocks into our final criterion for upgridding. This criterion has the capability to be modified to take more options into the account like cross flow. We introduced two new parameters, design factor and change of error per layer to find the optimum number of the layers without doing any further attempt to run the fine scale simulations. These new parameters can give some information about the extend of the heterogeneity in the reservoir and at the same time propose some optimum points for the coarse model that user can select according to his/her available time and memory. The out puts of the method also can give a qualitative idea that how bad or good the final coarse model will response. We presented several simple numerical cases to prove the advantages of the method over variation based methods and more we compared our results with King et. al method. We used this method for a real field case with 575 layers. We found the optimum number of the layer of 78 and compared our results with Kings and proportional upgrdding and proved that this new method gives a reasonable design and appropriate values for number of the optimum layers. References
1. Barker, J.W. and Thibeau, S., A Critical Review of the Use of Pseudorelative Permeabilities for Upscaling, SPE Reservoir Engineering, 138-143, May 1997. 2. King, M.J., Burn, K.S., Wang, P., Muralidharan, V., Alvardo, F., Ma, X. and Datta-Gupta, A., 2005. Optimal Coarsening of 3D Reservoir Models for Flow Simulation, SPE 95759, Annual Technical Conference and Exhibition, Dallas, Texas, U.S.A., 9-12 October 2005. 3. Li, D. and Beckner, B., Optimal Uplayering for Scaleup of Multimillion- Cell Geologic Models, SPE 62927, Annual Technical Conference and Exhibition, Dallas, Texas, U.S.A., 1-4 October 2000. 4. Testerman, J.D., A Statistical Reservoir-Zonation Technique, JPT, 889-893, August 1962. 5. Stern, D. and Dawson, A.G., A Technique for Generating Reservoir Simulation Grids to Preserve Geologic Heterogeneity, SPE 51942, Reservoir Simulation Symposium, Houston, Texas, U.S.A., 14-17 February 1999. 6. Zhang, P., Pickup, G. and Christie, M., A New Method for Accurate and Practical Upscaling in Highly Heterogeneous Reservoir Models, SPE 103760, International Oil & Gas Conference and Exhibition, Beijing, China, 5-7 December 2006. 7. Peaceman, D.W., Effective Transmissibilities of a Gridblock by Upscaling-Comparison of Direct Methods with Renormalization, SPE 36722, SPE Journal, 338-349, Volume 2, September 1997. 8. Li, D., Beckner, B. and Kumar, A., A New efficient Averaging Technique for Scaleup of Multimillion-Cell Geologic Models, SPE 56554, Annual Technical Conference and Exhibition, Houston, Texas, U.S.A., 3-6 October 1999. 9. Artus, V. and Noetinger, B.,Up-Scaling Two-Phase Flow in Heterogeneous Reservoirs: Current Trends, Oil & Gas Science and Technology-Rev. IFP, 185-195, Volume 59, No. 2, 2004. 10. Fisher, W.D., On Grouping for Maximum Homogeneity, Jour. Of Am. Stat. Assn., Volume 53, No. 284, 1958. 11. Fincham, A.E., Christensen, J.R., Barker, J.W. and Samier, P.,Up-Gridding from Geological Models to Simulation Model: Review, Applications and Limitations, SPE 90921, Annual Technical Conference and Exhibition, Houston, Texas, U.S.A., 2629 September 2004. 12. Meddaugh, W.S., Reservoir Modeling for Mature Fields-Impact of Work Flow and Upscaling on Fluid-Flow Response, SPE 99833, Europec/EAGE Annual Conference and Exhibition, Vienna, Austria, 12-15 June 2006.