Journal of Petroleum Science and Engineering 149 (2017) 707–719

Contents lists available at ScienceDirect

Journal of Petroleum Science and Engineering

journal homepage: www.elsevier.com/locate/petrol

Fast marching method assisted sector modeling: Application to simulation MARK

of giant reservoir models
⁎
Behzad Pouladia, Mohammad Sharifia, , Mohammad Ahmadia, Mohan Kelkarb
a
Department of Petroleum Engineering, Amirkabir University of Technology, Tehran, Iran
b
Williams Endowed Professor and Chairman, McDougall School of Petroleum Engineering, The University of Tulsa, 800 S Tucker Drive, Tulsa, OK 74104,
USA

A R T I C L E I N F O A BS T RAC T

Keywords: Primary geological model of reservoirs provided by geo-modeler usually contains complex structures and
Single well modeling millions of grid cells. Despite the recent advances in computational resources, these highly detailed models are
Drainage area not suitable for flow simulation purposes due to enormous simulation costs. Upscaling is necessary to reduce
Fast marching method the number of grid blocks in geological model while preserving the static and the dynamic properties of the fine
Streamline simulation
grid model. Upscaling is dependent on the flow process and it is often difficult to ensure that all the fine scale
details are preserved in the upscaling process. This adds to another level of uncertainty in dynamic performance.
Since so much effort has been put in constructing the detailed geological model, it is often desirable that the
simulation be carried out in the fine scale model.
This work proposes a new strategy regarding the simulation of fine grid models using a single well modeling
concept. The proposed method is currently suited for primary depletion process. The approach is relatively
straight forward. We first identify the drainage Volume associated with each well using Fast Marching method.
Fast Marching Method is extremely fast and can determine the drainage volume in large geological model within
minutes. We then assume that the drainage volume does not change significantly over the life of the well. We
then model each well separately as an independent well within a drainage volume and predict the performance
of the well. Because the simulation model size is significantly reduced, the well performance can be predicted
very quickly and is amenable to parallel simulations. By adding individual well performances, we can then
combine performance to predict the field performance. This method is also suitable for history matching since
by observing the drainage volumes of individual wells, we can reasonably determine if the model is suitable for
history matching before the history matching process starts.
We have tested the methodology for both the synthetic and the field cases and we have observed that the
method works well while significantly reducing the computational costs.

1. Introduction models. Upscaling is a remedy to this proposed problem. However, due

to difficulties associated with upscaling and the uncertainties that it
The ultimate goal of any field development and reservoir manage- adds to the new model, fine model simulation is preferable. This work
ment planning is to maximize recovery of hydrocarbons reservoirs. focuses on the simulation of fine grid models while keeping simulation
This requires meticulous and persistent analysis of the reservoir time reasonable. If we are able to isolate each well within its own
performance throughout the life of the reservoir. Reservoir perfor- drainage volume, the overall field simulation can be compartmenta-
mance is assessed and optimized by conducting simulation on a reliable lized into smaller simulations of individual wells. By assuming that the
reservoir model. The primary reservoir model is created by geo- drainage volume does not change significantly over the life of the field,
modeler which usually contains millions of grid cells. Simulation of we can simulate the entire reservoir performance very quickly. The
these fine grid cell models is not feasible because of significant concept of drainage volume is not new and it has been applied to
computational cost and time. CPU Time (Simulation time) in conven- optimize the reservoir performance as well as to determine the
tional simulation is roughly proportional to the square of the number of potential infill well locations. We apply the same concept to make the
grid cells (Thiele;1997). In spite of the advent of high power compu- overall simulation more efficient.
ters, it is still impractical to flow simulate high resolution reservoir The concept of drainage area (areal quantity of drainage volume)

⁎
Corresponding author.

http://dx.doi.org/10.1016/j.petrol.2016.11.011
Received 13 February 2016; Received in revised form 31 October 2016; Accepted 11 November 2016
Available online 15 November 2016
0920-4105/ © 2016 Elsevier B.V. All rights reserved.
B. Pouladi et al. Journal of Petroleum Science and Engineering 149 (2017) 707–719

Table 1 (1971) provided analytical solution for this type of test. He also showed
Values of constants for dimensionless formula of drainage radius. that shape of drainage area and well-location in the system can be
derived using the data from this test. In a paper published by Du
Author C1 C2
(2008) (Fay and Prats, 1951) a new approach based on integration of
Hurst et al. 2.6408 0.48658 conventional pressure transient analysis (PTA) and pressure rate de-
Van Poolen 2 1/2 convolution algorithm is presented which gives a generalization of
Jones 4 1/2
calculating drainage area and radius of investigation without any flow
regime presumption (radial flow). However, the well-testing analysis
has been around for a long time; however, even now, confusion gives only a quantity and fails to give a visual representation of the
abounds regarding the exact meaning of it. Hurst (1987) stated that drained area.
the remedy to recover the remaining oil in the reservoir is to determine In the current work, we calculate the drainage area of wells both
the area unaffected by the voidage. He proposed an approach by quantitatively and visually by using the Fast Marching Method (FMM).
streamline values based on Basset's formula. He also suggested a This method is faster compared to commonly proposed method based
formula for recovery index in terms of the difference between two on reservoir simulation and can also account for the heterogeneity and
successive streamline values in the numerator and the included area in connectivity of the reservoirs. The results are verified by streamlines
denominator. Samaniego et al. (1997) believed that the drainage area is generated by use of a streamline simulation.
strictly related to the drainage radius and suggested a formula for
radius of drainage based on the distance traveled during a specific 2. Theory
period by sonic velocity. He states that pressure gradient (and flow) can
occur only when pressure disturbance has been transmitted to a 2.1. Streamline simulation
specific reservoir volume. He then postulates that pressure disturbance
due to production at a well is propagated radially outward at local sonic Streamline simulation is an efficient alternative simulation techni-
velocity. Prior to Samaniego, several authors had addressed the que to traditional simulation methods which are based on the finite
estimation of radius of investigation and drainage area in terms of difference or finite volume discretization of the governing flow equa-
dimensionless time (Jones, 1962; Hurst et al., 1961; Van Poollen, tions. The earliest study dates back to the work of Muskat and Wyckoff
1964). All these formulations have the same dimensionless form as (1937) on stream function and the potential function in simple two-
rdD =C1 tD C2 where different authors have proposed different values for dimensional domains for incompressible flow. Roughly a decade later,
these constants as given in Table 1. A thorough review on various Fay and Prats (1951) were the first authors who reported its applica-
attempts on formulation of radius of investigation is given by Kuchuk tion in hydrocarbon reservoir simulation. A comprehensive historical
(2009). review on the development and the enhancement of this simulation
Samaniego finally presented an integrated method that combines technique, theory and detailed information on mathematical formula-
geological, seismic, well log, petro-physical and reservoir analysis tion is presented by Datta-Gupta and King (2007). Early streamline
pressure data to estimate the drainage area. Cox et al. (2005) methods were associated with simple fluid flow physics like incom-
conducted simulations to assess the parameters that affect the drainage pressible two-phase flow without gravity and capillary effects, but
area in tight gas wells. The variables they considered are effective gas recent modifications allow incorporation of more complex physics.
permeability, effective porosity, net thickness, reservoir geometry, Streamline simulation is employed for wide range of applications such
fracture stimulation half-length and the tubing pressure. They con- as history matching, reservoir management, up-scaling and ranking of
cluded that reservoir parameters, flow geometry, fracture half-length fine-scale geological models as well as a complementary tool to finite
and producing condition are important to determine the drainage area. difference simulation by providing important information about drai-
Dehdari et al. (2008) presented a methodology for appropriate well nage areas, flood surveillance and improvement of sweep efficiency (Al-
spacing and recovery optimization in one of the Iranian oil fields by Najem et al., 2012).
using streamline and traditional reservoir simulation. They utilized Streamlines are integrated curves that are tangential to a defined
numerical reservoir simulation to generate different property maps velocity field at a given instant of time (Al-Najem et al., 2012).
such as oil volume, water volume, permeability, phase saturations, etc. Streamline simulation is an IMPES simulation technique which means
They claimed that the best result for optimum location of the well is pressure equation is first solved implicitly and then saturation and
attained through simultaneous investigation of streamlines and prop- conservation equation are solved explicitly. By use of pressure dis-
erty maps. Al-amri et al. (2012) proposed a simple calculation method tribution, instantaneous velocity vectors perpendicular to pressure
for well drainage pressure and well drainage area. Their method is contours are generated. The velocity field is then utilized to map the
based on single down-hole pressure gauge, flowing well parameter and streamlines. Once the streamline is defined, the transport problem is
PVT data. Recently Gonzalez et al. (2014) published a method based on reduced into 1D problems over streamlines. The main advantages of
reservoir simulation in order to estimate drainage area and recovery streamline simulation can be attributed to three factors: (i) computa-
factor of an in-situ combustion project in a complex water-drive heavy tional speed and memory efficiency (Thiele et al., 2011) (ii) minimiza-
oil reservoir in Colombia. Their method comprises of generating the oil tion of numerical artifacts (iii) Visualization Potential (Al-Najem et al.,
drainage maps as a function of time. Drained-oil maps are determined 2012).
by subtracting the oil saturation at the simulation starting date from
the oil saturation at a specific date (time step). The key idea is to find a 2.2. Fast Marching Method (FMM)
boundary where the difference in the oil saturation is close to zero
which would be considered the influence area for in-situ combustion The level set method and Fast Marching Method (FMM) are two
project. They claimed that this method based on drained-oil maps is methods proposed to track the evolution of the interfaces (Sethian and
more accurate than pressure maps since strong aquifer acting in this Vladimirsky, 2000). These two methods are applied in wide variety of
reservoir does not allow sufficient pressure variation between wells. problems such as burning flames, wave's propagation in water, optical
In practice, a long drawdown test also called Reservoir Limit Test path planning, medical imaging, computational fluid dynamics, etc.
(RLT) is conducted to quantify the drainage area and consequently, the The two methods are different but complementary to each other. FMM
volume drained by a well (Affected Pore Volume), initial hydrocarbon can only be used for propagation problems in which the interface
in place, productive limits and deliverability (Jones, 1962). Earlougher strictly expands or contracts while the level set method is able to
account for propagation of interfaces that both expands and contracts.

708
B. Pouladi et al. Journal of Petroleum Science and Engineering 149 (2017) 707–719

FMM is closely related to Djikstra's method (Dijkstra, 1959) which 3.2. Calculation of drainage volume
allows computation of minimum-cost path on graph with different cost
attributed to the edge of the graph. Pressure propagation due to On the strength of previous subsection results, we infer that FMM is
production or injection by a well into the reservoir is a typical example able to represent physical and non-physical boundaries in reservoirs
of front movement in the porous media. and therefore enables us to determine drainage volume pertinent to
In the area of reservoir characterization, (Karlsen et al., 2000) first each well. Drainage volume by definition is the volume drained by each
presented a level set method for reservoir simulation based on well in the reservoir. For multi-well reservoirs, each well drains the
fractional flow formulation of two-phase, incompressible, immiscible blocks which are dynamically connected to the well's block. The
flow. They found the method to be effective and accurate for reservoir dynamic connectivity is well captured by FMM both qualitatively and
simulation, easy to program and avoids most of the numerical quantitatively. FMM is able to capture both physical and non-physical
difficulties associated with streamline methods. Lie and Juanes no flow boundary between wells and thus determining the drainage
(2005) presented a numerical simulation of first-contact miscible gas volume for each well. Stream line simulation is a well-known approach
injection using a front-tracking method. They assumed that the for drainage volume assessment in the literature. Streamlines for each
injection gas and the residual oil mix in all proportions to form a block start from the block and end to the one of the wells which shows
single hydrocarbon phase. Maša Prodanovic et al. (2010) investigated the blocks and their dynamically connected well. In plain English, they
the flow between fracture and adjacent matrix using a level set method allocate blocks to one of the wells. This can be used to calculate
for drainage and imbibition. Xie et al. (2012) presented a novel drainage volume. Results of streamline are used to validate results
approach to compute and visualize the depth of investigation in provided by FMM which showed FMM can capture drainage volume
unconventional reservoirs using FMM under very general reservoir for each well.
conditions. They also proposed a geometric pressure solution based on Authors have developed a computer code to record the blocks that
the depth of investigation, to estimate the transient pressure behavior host pressure drop front and are connected to pressure drop source
in unconventional wells with multistage fractures. Zhang et al. (2013) (well block). This allows for us to identify quantitatively the well
proposed a novel approach based on the (FMM) for performance drainage volume. Streamline simulation technique has been applied on
analysis of unconventional reservoirs with complex flow geometry and the same models to evaluate the accuracy of allocated drainage volume
anisotropic properties. They also derived the formulation of the FMM by FMM. In Case 1, a rectangular homogeneous gas reservoir with two
in corner point grids, which is one of the most widely used grids for symmetrically placed wells is considered. The model has 100×100×1
reservoir models. Recently Zhang et al. (2014) presented a novel blocks in x-, y- and z- direction respectively. Two wells are producing at
method for rapid simulation of the reservoirs based on FMM. This an equal rate of 100,000 ft^3/day. Other reservoir properties are given
method is based on a coordinate transformation from the physical in Table 2. The model is schematically shown in Fig. 2(a).
space to the one-dimensional space derived from the FMM solution. As one would expect, the two wells share the drainage volume of the
Their approach is analogous to the streamline method which is widely reservoir equally since all the parameters for both the wells are the
used for modeling convective flow in the reservoir. Sharifi et al. (2014) same. The predicted drainage volume by FMM and Streamlines are
presented a method based on FMM for ranking a large number of geo- qualitatively compared in (Fig. 2(b)).
cellular model realizations. They used FMM to quantify connected In Case 2, we apply the proposed approach to a heterogeneous oil
reservoir volume to a given well. They demonstrated that FMM is reservoir. Our heterogeneous isotropic model has 25×25×1 grids in x-,
computationally more efficient compared to both the flow and the y- and z- directions respectively. Our model has uniform porosity while
streamline simulation and can capture the dynamic connectivity of the the permeability distribution map is created using geo-statistical
reservoir. Sharifi and Kelkar (2014) applied this FMM for permeability simulation. The value of Dykstra-Parsons coefficient is 0.52 and
upscaling. They showed that the accuracy of upscaled permeability can average value of permeability is 6.82 md. Other model properties are
be evaluated by conducting FMM on both fine and coarse models. outlined in Table 3. Permeability distribution is shown in Fig. 3(a).
In the first example well locations are (11, 7) and (22, 21). As the
Fig. 3(b) shows, the result of both applied methods are in accordance
3. Results and discussion
with each other. Different well locations are also considered and results
are illustrated in Fig. 3(b).
In this subsection we present the application of FMM in determin-
Results show that FMM can give a satisfactory prediction for
ing the drainage volume. Next we consider the allocated drainage
drainage volume in just couple of seconds and thus, can be suitably
volume for each well by FMM as an independent reservoir with single
applied instead of streamline simulation. Grid independence and
well inside. First subsection shows the time map generated by FMM in
quantitative convergence study on value of drainage volume predicted
a homogeneous reservoir. Second subsection shows the ability of FMM
by streamline simulation and FMM is performed and provide in
to predict the drainage volume in homogeneous and heterogeneous
Appendix C. Case 3 is a 60×130×1 (dimensions in x-, y- and z-
reservoirs. In third subsection we apply the concept of single well
direction respectively) model borrowed from the first layer of SPE 10
modeling using the predetermined drainage volume predicted by
comparative model. The model description can be readily found in the
FMM.
literature. Three wells are randomly distributed in the reservoir.
Drainage volume for the three wells are determined using stream-
3.1. Time map in homogeneous reservoir line and FMM (Fig. 4(a)) and are graphically compared in Fig. 4(b).
This figure shows that the drainage volume predicted by FMM is in
FMM calculates the arrival time of front propagation for every point accordance to the drainage volume occupied by streamlines.
in reservoir with respect to an origin (starting point) and gives a time
map as depicted in Fig. 1(a). This figure also shows the investigated 3.3. Single well modeling
volume by pressure drop front at a specific time.
FMM can also be applied to reservoirs with more than one well. It is The key idea behind single well modeling is segmentation of whole
also applicable to problems in which wells start operating at different reservoir domain into several discrete drainage volumes for all the
times. A time map for a homogeneous reservoir with two wells wells in the reservoir and then independently simulating each drainage
symmetrically located within the reservoir is shown in Fig. 1(b). It volume with a single well. Single well modeling is different from the
can be clearly seen that no-flow boundary between two wells is well parallel option offered by the simulators and can more effectively
captured by FMM and visible in time map. reduce the computational cost since it avoids most of the complexities

709
B. Pouladi et al. Journal of Petroleum Science and Engineering 149 (2017) 707–719

Fig. 1. (a): Time Map Generated for a Single Well in a Homogeneous Reservoir. (b): Time Map Generated for Homogeneous Reservoir with Two Wells.

Table 2 Table 3
Homogeneous reservoir properties. Heterogeneous model properties.

Reservoir properties Case1- homogeneous Reservoir properties Case2- heterogeneous

^3
Grid block size 20×20×100 ft Grid block size 50×50×250 ft^3
Porosity 0.1 Porosity 0.25
Permeability 1 md Compressibility 1E−6 1/psi
Compressibility 2.0E-5 1/psi Viscosity 6.82 cp
Viscosity 0.0314 cp Initial Pressure 5793.6 psi

and time-consuming processes associated with parallel simulation. The discussed in the previous section. In subsection 3 we apply single well
work flow for the single well modeling is as follows: modeling to the model built on the basis of first layer of SPE10
comparative model. We finally present application of single well
• Apply FMM to determine the drainage volume for every active well modeling to a real gas condensate field in the Middle East.
in the reservoir.
• Create single well model for a specific well by deactivating reservoir 3.3.1. Single well modeling- homogeneous reservoir
blocks that do not lie in well's assigned drainage volume. In this subsection, we apply the single well modeling approach to
• Simulate all created single well models individually the homogeneous reservoir model previously discussed. As previously
stated, drainage volume is equally shared between two wells in the
Authors believe that determined drainage volume for each well considered homogeneous reservoir.
remains unchanged unless production constraint changes or new infill Performance of Single Well Modeling in a homogeneous reservoir is
wells are drilled. In the current section we use the drainage volume analyzed by comparing the obtained results with those of full reservoir
determined earlier for homogeneous and heterogeneous reservoirs simulation. Fig. 5 shows the bottomhole pressure for well 1 and 2
examined in previous section. Then we simulate each well with its predicted by single well model and the result obtained from full
associated drainage volume as a new independent reservoir with a reservoir simulation.
single well inside. In subsection 1 we apply the single well modeling to
the considered homogeneous reservoir. In subsection 2, single well
modeling performance is assessed in the heterogeneous reservoir 3.3.2. Single well modeling- synthetic heterogeneous reservoir
In the following, we employ the single well modeling to the

Fig. 2. (a): Schematic view of homogeneous reservoir model. (b): Qualitative comparison of drainage volume calculated by FMM and streamlines.

710
B. Pouladi et al. Journal of Petroleum Science and Engineering 149 (2017) 707–719

Fig. 3. (a): Permeability Distribution – Synthetic Heterogeneous Reservoir Model. (b): Drainage Volume Predicted by FMM, Streamline Simulation and their Qualitative Comparison –
Synthetic Heterogeneous Reservoir Model.

711
B. Pouladi et al. Journal of Petroleum Science and Engineering 149 (2017) 707–719

Fig. 4. (a): Areal view of drainage volume predicted by FMM- First Layer of SPE 10. (b): Qualitative comparision of drainage volume predicted by streamline and FMM- First Layer of
SPE 10.

Fig. 5. Bottom hole pressure predicted by singel well model and full reservoir simulation
for Well 1 and Well 2- Homogeneous Reservoir Model.

heterogeneous reservoir that was previously considered in subsection

3.2. For the first example illustrated in subsection 3.2, we apply the
single well modeling using the drainage volume predicted by FMM.
Then we compare the results obtained by single well modeling with the
results obtained from full reservoir simulation. Fig. 6(a) illustrates
drainage volume used for well 1 in single well modeling. Results
comparing performance of a single well model and full reservoir
simulation for well 1 and 2 are given in Fig. 6(b).
We observe that bottomhole pressures predicted by single well
modeling are in good agreement with those of full reservoir simulation.

3.3.3. Single well modeling- first layer of SPE10 model Fig. 6. (a): Single Well Model used for Well 1- Synthetic Heterogenous Reservoir Model.
Now we apply the single well modeling approach to the model built (b): Bottom Hole Pressure Predicted by Singel Well Model and Full Reservoir Simulation
on the basis of first layer of SPE 10 comparative model. We employ the for Well 1 and 2- Synthetic Heterogeneous Reservoir Model.
predicted drainage volume by FMM to construct the single well models.
Fig. 7(a) shows the drainage volume and single well model for each well Fig. 8(a). Maximum value of porosity is 0.34, minimum value is 0.04
in 3D. while maximum and minimum values of permeability are 75 md and 2
In order to assess the accuracy of the approach, bottomhole md respectively. Average values for porosity and permeability are 0.15
pressure obtained by simulation of single well models are compared and 7.48 md respectively, and the calculated standard deviation for
with those obtained when the entire reservoir is simulated. Fig. 7(b), porosity and permeability are 0.067 and 7.7285 respectively. Nine
shows the bottomhole pressure comparison for 15 years of simulation producing wells are located across the model and producing with a
for well 1, well 2 and well 3. constant gas rate of 100 MMSCF/day. Wells are perforated in all 19
The match between bottomhole pressure predicted by single well producing layers. The reservoir fluid components are lumped into
modeling and those of full reservoir simulation is satisfactory, espe- following 7 groups, H2S, CO2, N2C1, C2-C3, C4-C6, C7-C15 and C16+
cially in early years. where the individual molar fractions are 0.0024, 0.0193, 0.861, 0.072,
0.0228, 0.0199 and 0.0026. The Peng-Robinson equation of state is
3.3.4. Single well modeling- real field application used to model fluid properties. The initial water saturation is 0.25 and
In this subsection, we present an example in which we apply the the relative permeability data is provided in Fig. 8(b).
single well modeling to a real gas condensate field in the Middle East. We applied the FMM to find the drainage volume for all nine
The model contains 129×92×30 grid blocks in x, y and z directions producing wells in order to build the single well model for all the nine
respectively. Porosity and permeability distribution is depicted in wells. Predicted drainage volumes are shown in Fig. 8(c). Fig. 8(c) also

712
B. Pouladi et al. Journal of Petroleum Science and Engineering 149 (2017) 707–719

FMM to determine the wells drainage volumes are compared. We

compare time efficiency of FMM with the streamline simulation using a
very heterogeneous reservoir that has two producing wells. CPU time
calculated for both methods is compared in Fig. 9.
It can be clearly deduced that FMM is orders of magnitude faster
than streamline simulation. This is actually due to the fact that
streamline simulation approach solves flow equations for all grid
blocks while FMM only calculates the speed function, which is a very
simple expression and then calculates the arrival time through an easy
to implement algorithm. We have also employed the heap structure in
our computer code to further enhance the time efficiency of FMM.
Description on heap structure can be found in Yatziv et al. (2006).

3.5. Application and limitation of single well modeling approach

As discussed in this research, the main application of this method is

facilitating the simulation of giant reservoirs during primary depletion.
While there exist no simple documented method in literature to
determine drainage volume of each well without need for flow
simulation, drainage volume can be calculated very quickly and
accurately using the current approach. Then instead of focusing on
whole reservoir, one can focus on single well models and try to
investigate within drainage volume of each well. We showed that for
depletion production, even for heterogeneous and field cases, dynamic
response of the reservoir can be captured reasonably using single well
modeling rather than conducting flow simulation on full field which
can be used in history matching process. We can focus on each wells
individually and individual well response (i.e. DST, production) can be
matched much easier.
Use of this methodology is limited to the periods in which
production occurs under constraints which do not change significantly
and with definable initial and boundary conditions. It should be also
emphasized that the proposed method will not work if the field is
subjected to water flooding since the process becomes convection
dominated. However, if the field is producing under primary depletion
(expansion), the results can be consistent and accurate.

Fig. 7. (a): Drainage Volume depicted for 3 wells and used for single well modeling-
First Layer of SPE 10 Reservoir Model. (b): Bottom hole pressure predicted by singel well 4. Conclusions
model and full reservoir simulation for Well 1, 2 and 3- First Layer of SPE 10 Reservoir
Model. We have presented a novel approach that can be effectively used to
simulate, large scale, fine grid models. This approach consists of
shows the created single well model for well 4. splitting the whole reservoir model into several sub-models each
In the Fig. 8(d), a detailed comparison between bottomhole comprising the drainage volume of a single well. Each sub-model is
pressure for 15 years of simulation predicted by simulating single well then simulated independently. The wellbore behavior matches per-
models and full reservoir simulation is provided. In these figures, we fectly with the results obtained from the full field reservoir simulation.
show the two best (wells 4 and 8) and the two worst prediction (wells 9 The method is limited to simple depletion processes and can be applied
and 2). Also, the final value of bottomhole pressure and the deviation when the number of wells is constant. It is ideally suited during the
for single well approach and full reservoir simulation for all wells are early stages of the field life where most wells are producing under
summarized in Table 4. We have also calculated the difference between depletion drive. History matching using single wells would be much
bottomhole pressure predicted by single well modeling and full more efficient than using multiple well production data.
reservoir simulation at the final time step which represents the Compared to streamline simulation, FMM method is computation-
maximum error. ally much more efficient. This is because of relatively simple, algebraic
We observe negligible deviation and satisfactory agreement be- and local nature of FMM method. We applied the FMM to calculate
tween bottomhole pressures predicted by single well modeling ap- drainage volume for each well without conducting flow simulation.
proach and those predicted by full reservoir simulation after 15 years of FMM reduces the time required to predict drainage volume for fine grid
simulation. The simulation time required for single well simulation was models to few seconds with enough accuracy. Our approach is applied
7 times smaller than required for full field simulation. to multiple synthetic and field cases. In all cases, drainage volume
predicted by FMM is consistent with those obtained from streamline
3.4. Time efficiency simulation. Dynamic response (i.e.; bottomhole pressure) obtained
from running each individual well model is also consistent with the
In this subsection, the time required by streamline simulation and entire reservoir model.

713
B. Pouladi et al. Journal of Petroleum Science and Engineering 149 (2017) 707–719

Fig. 8. (a): Porosity and permeability distribution- real field reservoir model. (b): Relative permeability data for gas-water and oil-water systems. (c): Drainage volume predicted for nine
wells by FMM and single well model used for well 4. (d): Bottom hole pressure predicted by singel well model and full reservoir simulation for well 1, 2 ,3 and 4- real field reservoir
model.

714
B. Pouladi et al. Journal of Petroleum Science and Engineering 149 (2017) 707–719

Table 4
Bottomhole pressure predicted by single well models and full reservoir simulation after 15 years of production.

Well number 1 2 3 4 5 6 7 8 9

Full -BHP (psi) 4684.25 4736.91 4731.34 4651.4 4631.46 4700.46 4702.52 4679.78 4801.96
Single-BHP (psi) 4672.79 4724.78 4735.76 4648.7 4625.2 4685.82 4705.2 4679.04 4823.19
Error (psi) 11.46 12.23 4.42 2.7 6.26 14.46 2.86 0.74 21.23

Fig. 9. Comparision of CPU time for streamline simulation and FMM.

Appendix A. Formulation of interface propagation problem

The pressure transient response in a heterogeneous porous media is governed by well-known diffusivity equation
∂P
ϕ (x ) μct = ∇. (k (x )∇P (x, t ))
∂t (1)
Where P(x,t) represents the pressure, ϕ (x ) denotes the porosity, k (x )denotes the permeability, μ and ct are fluid viscosity and total compressibility
respectively. A Fourier Transform of Eq. (1) results in the following equation in the frequency domain.

˘P (x, ω) = 21π ∫ P (x, t ) e −iωt dω

(2)

μct (x ) φ (x ) ∇ k (x )
k (x )
˘
iwP (x, ω) −
k (x )
˘ ˘
. ∇P (x , ω) − ∇2 P (x, ω) = 0
(3)
Assuming that ∇ k (x )
is negligible, the final form of above equation is
k (x )
1
˘
∇2 P (x, ω) −
η (x )
iωP = 0 ˘ (4)
k (x )
Where η (x ) = μct (x ) φ (x )
Asymptotic approach (Vasco et al., 2000) follows if we consider a solution in terms of inverse power of −iω
∞
Ak (x )
˘P (x, ω) = e − −iω τ (x )
∑
k =0 ( −iω )
k
(5)
Where ω is the frequency, τ (x ) is the propagation time of pressure front also diffusive time of flight and Ak (x ) is the pressure amplitude at kth order.
The advantage of this form of solution is the dominancy of initial terms of series at high frequency. In this study we are interested in front location,
therefore propagation of sharp front of pressure can be described by initial terms (Vasco et al., 2000). The high frequency solution given by initial
terms of asymptotic series corresponds to following form of solution

P̆ (x, ω) = A0 (x ) e− −iω τ (x )
(6)
2
Inserting Eq. (6) into Eq. (4) and collecting the terms of the highest order in −iω , that is ( −iω ) results in the following well-known Eikonal
equation for propagating front.

715
B. Pouladi et al. Journal of Petroleum Science and Engineering 149 (2017) 707–719

α (x ) ∇ τ (x ) = 1 (7)
k (x )
where α = μct (x ) φ (x )
The above equation simply tells that the pressure front propagates in the reservoir with the velocity equal to square root of diffusivity which is
independent of flow rate and only dependent on porous media and fluid properties. It is also important to note that the assumption made in
derivation of Eq. (4) was necessary in order to achieve the form of Eikonal equation which enable us to apply FMM to solve the equation.
The time of flight can be physically interpreted to real time domain by use of inverse Fourier Transform. Kulkarni et al. (2001) showed that the
transformation of aforementioned equation in time domain is

τ (x ) ⎡ τ 2 (x ) ⎤
P (x , t ) = A 0 (x ) exp ⎢ − ⎥
2 πt ⎣ 4t ⎦ (8)
However, this form of solution is still not convenient to use. We can calculate the time needed for a pressure propagation to reach its maximum
or minimum by taking derivative of dP (x, t ) to zero, i.e. the maximum time that the pressure response at point x reaches its peak value. Taking the
dt
derivative leads to following relation between time and diffusive time of flight.
τ 2 (x )
tmax =
4 (9)
Diffusive time of flight τ (x ) has dimension of square root of time and closely related to the concept of radius of investigation in reservoir
engineering. In general the diffusive time of flight is related to real time domain by following general relation given as follows:
τ 2 (x )
tmax =
c (10)
where c is a pre-factor dependent on the flow geometry. The value of pre-factor for linear flow geometry is c=2 whereas the value for radial flow
and spherical flow is c=4 and c=6 respectively.

Appendix B. Solution to interface propagation problem using FMM

FMM is a front tracking method that can efficiently solve the Eikonal equation. The general form of Eikonal Equation is as follows. Diffusivity
equation in porous media can be written in the form of Eikonal equation (Xie et al., 2012). Interested readers are encouraged to see Appendix A for
details on derivation of Eikonal equation in porous media.
F (x ) ∇ τ (x ) = 1 (2.1)
This method calculates the arrival time of pressure front at a specific location as long as the velocity function (F(x)) is known. In Appendix A, we
derived the pressure drop front propagation in porous media in the form of Eikonal equation and we showed that the velocity function is dependent
on porous media and fluid properties at that location. In Cartesian coordinate, by knowing the value of fluid and porous media properties at position
of (x,y,z) and setting the arrival time (τ (x )) at starting point to zero (boundary condition for eikonal equation), we may find the arrival time at other
locations by solving the derived Eikonal equation using FMM. It is worth mentioning that this method is not restricted to Cartesian coordinate and
can be also applied to triangular (Barth and Sethian, 1998) and unstructured (Hysing and Turek, 2005) gridding schemes. However, for simplicity
in our application we discuss the Cartesian gridding.
We illustrate the solution to Eikonal equation on the rectangular grid by using Finite Difference approximation. Considering that the domain is
discretized by orthogonal rectangular mesh to following block size Δx (in x-direction), Δy (in y-direction) and Δz (in z-direction), the first order fast
marching formulation in 3D to calculate the arrival time at position (x,y,z) is as follows:
τ = τi, j, k
τx = min(τi −1, j, k , τi +1, j, k )
τy = min(τi, j −1, k , τi, j +1, k )
τz = min(τi, j, k −1, τi, j, k +1) (2.2)

⎛ τ − τx ⎞2 ⎛ τ − τy ⎞2 ⎛ τ − τz ⎞2 1
max ⎜ , 0⎟ + max ⎜ , 0⎟ + max ⎜ , 0⎟ = 2
⎝ Δx ⎠ ⎝ Δy ⎠ ⎝ Δz ⎠ F (2.3)
Where i, j, k are the indices in the x, y and z-directions respectively. In order to apply the FMM to petroleum reservoirs, we first discretize the
reservoir domain to desired block sizes, next we choose a starting point where the arrival time is zero and then we begin calculating arrival time for
neighboring blocks until we reach the boundary of the drainage volume. Actually reservoir domain is divided into following three block sets: (i)
Frozen points set (The blocks for which their arrival times are already computed) (ii) Narrow band set (The blocks that are possible candidates to
host pressure front) (iii) Far away points set (The blocks which are not in the neighborhood of pressure front and are not considered for arrival time
calculation). The algorithm for finding the arrival time for a desired block works as follows and schematically shown in the figure below:

1. Choose Starting Point. Arrival time (τ ) for starting point is zero. In petroleum engineering application, well locations are chosen as the starting
points. This is physically meaningful since the pressure drop front starts propagating from the well blocks. Add the starting point to frozen set.
2. Find non-frozen neighbor blocks of frozen points. Add the neighbor blocks to narrow band set.
3. Calculate the arrival time for blocks in Narrow band set using Eqs. (2.2)–(2.3).
4. Find the block with the lowest value of arrival time (τ ).
5. Add the block to Frozen set and remove it from Narrow band set.
6. Repeat steps 2–5 until all points become frozen (Arrival time is calculated for all points).

We note that if the arrival time for some blocks required for computing τx ,τy or τz in Eq. (2.2) is not provided, then we ignore the terms in Eq.

716
B. Pouladi et al. Journal of Petroleum Science and Engineering 149 (2017) 707–719

Fig. B1. Fast marching method algorithm implementation.

(2.3) that contain these unknowns.

It is worth pointing out that the general first order fast marching formula in 3D can also be reduced to 2D problems by simply neglecting the
terms in Eqs. (2.2) and (2.3) that are associated with intended dimension (Fig. B1).

Appendix C. Grid independence and quantitative convergence study on the drainage volume predicted by streamline
simulation and FMM

The synthetic heterogeneous reservoir model used in Section 3.2 is considered. The original reservoir dimension is 25× 25. In order to assess the
effect of grid refinement on predicted drainage volume, the reservoir is refined to 50× 50 and 100× 100 grid dimensions. Quantitative and qualitative
comparison are given in the following. Results reveal that grid refinement doesn’t alter the predicted drainage significantly neither in streamline
simulation approach nor in Fast Marching Method approach. We can also infer that drainage volume predicted by streamline simulation and Fast
Marching Method tend to the same drainage volume partition (Fig. C1).

Drainage Volume Predicted by FMM (%) Drainage Volume Predicted by Streamline simulation (%)

Model size Well 1 Well 2 Well 1 Well 2

25× 25 39.68 60.32 41.64 58.32

50× 50 40.96 59.04 42.19 57.81
100× 100 41.36 58.64 42.78 57.13

717
B. Pouladi et al. Journal of Petroleum Science and Engineering 149 (2017) 707–719

Fig. C1. Grid Independency Test.

References reservoir simulation to estimate drainage area and recovery factor of an in-situ
combustion project in a complex water-drive heavy oil reservoir. In: Proceedings of
the SPE Heavy Oil Conference-Canada, Society of Petroleum Engineers.
Al-amri, M.A., Al-Khelaiwi, F.T., Kadem, M., 2012. Real-time estimation of well drainage Hurst, W., 1987. Drainage and In-Fill Drilling.
parameters. In: Proceedings of the SPE Kuwait International Petroleum Conference Hurst, W., Haynie, O.K., Christie, M., Walker, R.N., 1961. Some problems in pressure
and Exhibition, Society of Petroleum Engineers. build-up. In: Proceedings of the Fall Meeting of the Society of Petroleum Engineers
Al-Najem, A., Siddiqui, S., Soliman, M., Yuen, B., 2012. Streamline simulation of AIME, Society of Petroleum Engineers.
technology: evolution and recent trends. In: Proceedings of the SPE Saudi Arabia Hysing, S.-R., Turek, S., 2005. In The Eikonal equation: numerical efficiency vs.
Section Technical Symposium and Exhibition, Society of Petroleum Engineers. algorithmic complexity on quadrilateral grids. In: Proceedings of ALGORITMY. pp
Barth, T.J., Sethian, J.A., 1998. Numerical schemes for the Hamilton–Jacobi and level 22–31
set equations on triangulated domains. J. Comput. Phys. 145 (1), 1–40. Jones, P., 1962. Reservoir limit test on gas wells. J. Pet. Technol. 14 (06), 613–619.
Cox, S.A., Sutton, R.P., Stoltz, R.P., Knobloch, T.S., 2005. Determination of effective Karlsen, K.H., Lie, K.-A., Risebro, N., 2000. A fast marching method for reservoir
drainage area for tight gas wells, In: Proceedings of the SPE Eastern Regional simulation. Comput. Geosci. 4 (2), 185–206.
Meeting, Society of Petroleum Engineers. Kuchuk, F.J., 2009. SPE 120515 Radius of Investigation for Reserve Estimation from
Datta-Gupta, A., King, M.J., 2007. Streamline Simulation: Theory and Practice. Society Pressure Transient Well Tests.
of Petroleum Engineers, Richardson, TX. Kulkarni, K.N., Datta-Gupta, A., Vasco, D.W., 2001. A streamline approach for
Dehdari, V., Aminshahidy, B., Tabatabaei-nejad, A., 2008. Well spacing and recovery integrating transient pressure data into high-resolution reservoir models. SPE J. 6
optimization of one of iranian oil fields by using streamline and reservoir simulation. (03), 273–282.
In: Proceedings of the Paper SPE 112985 Presented at the SPE Western Regional Lie, K.-A., Juanes, R., 2005. A front-tracking method for the simulation of three-phase
and Pacific Section AAPG Joint Meeting, Bakersfield, California, USA, 29 March-2 flow in porous media. Comput. Geosci. 9 (1), 29–59.
April. Maša Prodanovic, S., Bryant, S.L., Karpyn, Z.T., 2010. Investigating matrix/fracture
Dijkstra, E.W., 1959. A note on two problems in connexion with graphs. Numer. Math. 1 transfer via a level set method for drainage and imbibition.
(1), 269–271. Muskat, M., Wyckoff, R.D., 1937. The Flow of Homogeneous Fluids Through Porous
Du, K., 2008. The determination of tested drainage area and reservoir characterisation Media 12. McGraw-Hill, New York.
from entire well-test history by deconvolution and conventional pressure-transient Samaniego, F., Montiel, D., Perez, G., Arriola, A., 1997. [RFP] 6 New techniques for the
analysis techniques. In: Proceedings of the SPE Annual Technical Conference and assessment of drainage areas and reservoir dynamics. In: Proceedings of the 15th
Exhibition, Society of Petroleum Engineers. World Petroleum Congress, World Petroleum Congress.
Earlougher, R., Jr, 1971. Estimating drainage shapes from reservoir limit tests. J. Pet. Sethian, J.A., Vladimirsky, A., 2000. Fast methods for the Eikonal and related Hamilton–
Technol. 23 (10), 1266–1268. Jacobi equations on unstructured meshes. Proc. Natl. Acad. Sci. 97 (11), 5699–5703.
Fay, C., Prats, M., 1951. The application of numerical methods to cycling and flooding Sharifi, M., Kelkar, 2014. Novel permeability upscaling method using fast marching
problems. In: Proceedings of the 3rd World Petroleum Congress. pp 101–112 method. Fuel 117 (Part A), 568–578.
Gonzalez, L., Ferrer, J., Fuenmayor, M., Castillo, N., Gil, E., Farouq Ali, S., 2014. Use of Sharifi, M., Kelkar, M., Bahar, A., Slettebo, T., 2014. Dynamic ranking of multiple

718
B. Pouladi et al. Journal of Petroleum Science and Engineering 149 (2017) 707–719

realizations by use of the fast-marching method. SPEJ.. Proceedings of the SPE Europec/EAGE Annual Conference, Society of Petroleum
Thiele, M.R., Batycky, R.P., Blunt, M.J., 1997. A streamline-based 3d field-scale Engineers.
compositional reservoir simulator. In: Proceedings of the SPE Annual Technical Yatziv, L., Bartesaghim, A., Sapiro, G., 2006. O(N) Implementation of the Fast Marching
Conference and Exhibition, Society of Petroleum Engineers. Algorithm, J. Computational Physics, vol. 212, pp. 393-399.
Thiele, M.R., Gerritsen, M.G., Blunt, M.J., 2011. Streamline simulation. Soc. Pet. Eng.. Zhang, Y., Yang, C., King, M.J., Datta-Gupta, A., 2013. Fast-marching methods for
Van Poollen, H., 1964. Radius-of-drainage and stabilization-time equations. Oil Gas. J. complex grids and anisotropic permeabilities: application to unconventional
62 (37), 138–146. reservoirs. In: SPE Reservoir Simulation Symposium, Society of Petroleum
Vasco, D., Keers, H., Karasaki, K., 2000. Estimation of reservoir properties using Engineers.
transient pressure data: an asymptotic approach. Water Resour. Res. 36 (12), Zhang, Y., Bansal, N., Fujita, Y., Datta-gupta, A., King, M.J., Sankaran, S., 2014. From
3447–3465. streamlines to fast marching: rapid simulation and performance assessment of shale
Xie, J., Gupta, N., King, M.J., Datta-Gupta, A., 2012. Depth of investigation and gas reservoirs using diffusive time of flight as a spatial coordinate. In: Proceedings of
depletion behavior in unconventional reservoirs using fast marching methods. In: the SPE Unconventional Resources Conference, Society of Petroleum Engineers.

719