Sei sulla pagina 1di 22

SPE-195798-MS

Hybrid Methods for Analysis of Fractured Well Production from Liquids Rich
Duvernay Shale

Jagannathan Mahadevan and Huanzhen Hu, University of Southern California

Copyright 2019, Society of Petroleum Engineers

This paper was prepared for presentation at the SPE Annual Technical Conference and Exhibition held in Calgary, Alberta, Canada, 30 Sep - 2 October 2019.

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
Objectives/Scope: In order to maximize the recovery of hydrocarbons from liquids rich shale reservoir
systems, the cause and effect relationships between production and the stimulation methods need to be
clearly understood. In this study, we integrate a production data regression approach with flow simulation
methods to understand the fractured well production behavior and field wide well performance in a liquids
rich petroleum system in the Duvernay Basin.
Methods, Procedures, Process: Statistical models assume no physical relationship between the model
parameters and the response variable, which in this case is produced volumes over a period of time. On the
other hand, simulation studies incorporate physical mechanisms of flow to model and predict the production
behavior. The simulation models, however, fall short of incorporating all the mechanisms contributing to
the production behavior in the complex shale gas reservoir. Thus there is a need for integration of statistical
approaches of understanding production behavior along with physics based model and simulation approach.
Results, Observations, Conclusions: Multivariate linear regression analysis of the 6 month produced
volume and its relationship with parameters such as fracture fluid volumes used, proppant weight placed,
and number of stages fractured provides a model with reasonably good correlation. The 6 month produced
volumes correlate with large proppant weights, lower fluid placements and greater density of fracture
stages. Use of Random Forests machine learning algorithm on the dataset confirms that the total proppant
placed, well length completed with fractures have high importance coefficients. In order to examine the
well performance using full physical models, fractured well simulations were performed on particular wells
using the trilinear model. The trilinear model predictions were compared against other production analyses
and the regression model results for consistency. The models showed that in the absence of stress dependent
permeability, the production forecast was much higher. Thus, stress dependent permeability appears to be
an important factor in the modeling and prediction of production from liquids rich shale reservoirs.
Novel/Additive Information: In this study we describe a method to understand the production data
from a liquids rich shale reservoir, by integrating multivariate linear regression analysis, machine learning
algorithms along with physical model simulations. The results are novel and offer a method to validate
either approach to understand cause and effect relationships. This approach may be classified as a new
2 SPE-195798-MS

hybrid modeling approach that may potentially be used to optimize stimulation techniques in liquids rich
shale reservoirs.

Introduction
The prediction of well and reservoir performance in unconventional systems is of importance to almost all
levels of a development project. Usually pilot wells, and their production characteristics obtained through
pressure transient testing, is available to understand the far field properites of the reservoir. However, in
unconventional wells, the pressure transient testing is usually limited to very near the fracture or the well
itself. To attain a radial or pseudo radial flow condition in a Thus there is very little that can be gained to
understand the far field unless the wells have produced for a very long time.
A good underdstanding of the reservoir is very important in developing predicitive capabilities and hence
the usual approach is to develop mathematical models which are then constrained with known data that vary
with time, such as pressure and rates of production of fluids. Clarkson and William-Kovacs, 2013, discuss
modeling of 2-phase flow at early production periods including fracture fluid load recovery response. The
article discusses various flow regimes in the course of flowback and production. The early time flowback
data is used to obtain the hydraulic fracture properties such as fracture permeability and fracture drainage
radius which is compared to the fracture half-length from rate-transient analysis of the production data. The
approach is nevertheless a full-physics approach.
A comparatively longer time period evaluation of the unconventional well production is presented in
Clarkson et al. (2013) which considers time dependency of porosity and permeability of both matrix and
fracture during production in their model and spans several years of production history. The approach is
analytical including the physics of flow process. The study concludes that the degradation in the fracture
conductivity and matrix permeability with decline in average reservoir pressure could be a permanent and
irreversible phenomena.
Because the exact physical mechanisms of flow process is complex it is convenient to cast the problem
into a cause and effect relationship exercise using a statistical approach. The various input parameters are
also called as the predictors and the desired factor that concerns the practioner is the response. In using
these terms we adopt from the study of Vyas et al. (2017). The weight or coefficients of the factors may be
obtained by a regression of the model to the data.
The cause and effect relationships made through a simulation model can be either include all the known
physical mechanisms or can be semi-empirical. The physical models require constraining to the data by the
way of history matching and parameter estimation. Application of machine learning techniques have gained
prevalence in the petroleum industry with several new approaches and workflows ranging from recovery
forecasting to completion optimization. Using simple empirical models for well declines are compiled
and analyzed with machine learning algorithms to understand the impact of the completion parameters on
response variables such as initial flow rates (Vyas et al., 2017). The Vyas et al. 2017 study uses algorithms
such as Random Forests Network regression method, Support Vector Machines and Multivariate Adaptive
Regression Spilnes (MARS) to model the well decline.
In this study we use machine learning methods including Random Forests Network regression algorithm
and multiple linear regression algorithm to understand the impact of various predictors on a given response.
We use the results from the regression modeling to narrow down the range of parameters that will be
important in the full-physics model that is to follow. The resulting model is then history matched with the
production data with a greater level of confidence in the uniqueness of the history match results.
SPE-195798-MS 3

Duvernay Liquids Rich Formation Well Performance


Geologic Description
We consider production from two subgroup of wells in the large Duvernay formation. One subgroup Group
1 is located in proximity to the rich gas condensate rim as described in Figure 1. Group 1 unconventional
reservoir in the Western Canada Sedimentary Basin (WCSB) is a large elongate feature that consists of a
massive wedge of sedimentary rock extending from the Rocky Mountains in the west to the Canadian Shield
in the east. It comprises the eastern Canadian Cordillera and two sub basins: Alberta Basin and Williston
Basin. Geologically Group 1 area belongs to the Alberta sub basin of the WCSB. The three main formations
considered are Ireton, Upper Duvernay and the Lower Duvernay which are separated by a layer known as
"Middle Carbonates." The second subgroup, Group 2, is a much larger well group located north-west of
the Group 1 wells.

Figure 1—The schematic on the left shows a representation of areal extent of the western Canada sedimentary basin.
Schematic on the right shows a representation of the areal distribution of fluid properties, with dry gas to the west
and condensate rich system towards the east. The wells modeled in this study are marked on this schematic by a star.

The reservoir geology and the properties are broadly discussed in the previous section. From the point
of view of the reservoir engineering, it will be important to understand the nature of reservoir fluids, their
relative contents, the properties of the reservoir rock and the interaction between the fluids and the rock
during both fracturing and production. Fracturing operation, to create fracture surface area and keep the
fractures open for flow of reservoir fluids, is described later in the report.
The fracturing operation and the related geomechanical alterations to create access or surface area is the
most important engineering component in exploiting the resources in Duvernay. The fracturing operation
exposes existing faults and fractures, as well as creates new fractures that provide pathways for the injected
fluids to flow. The expectation is that the injected fluids will carry the proppant and keep the fracture open
for flow of the reservoir fluids in to the fracture and then into the well.

Well Description
The wells considered in study are placed in the Upper Duvernay formation and the multiple stage hydraulic
fractures are assumed to span the entire thickness of the formation. The first Well A is located in a pad
slightly away from the rich gas window and is the highest producing well from the production analysis. This
4 SPE-195798-MS

well was stimulated utilizing a Slickwater fracturing treatment in a 30 stage interval. Each stage includes
4 clusters and 160 shots were performed for each stage.
The data for the simulation was calibrated based on available data and assumptions. A pressure match
was performed to determine likely fracture geometry within the fracture model. The saturation pressure of
the reservoir fluid was determined to decrease from west to east, as is expected with volatile oil trending
to the east and dry gas trending to the west.
The thermodynamic phase behavior of the hydrocarbon fluids is of critical importance in the modeling
and analysis of production data. The behavior of the produced volumes averaged over 6 month period is a
strong function of the location and depth of the well.

Matching Historical Data with Physical Models


Trilinear Model Description

Figure 2—Schematic of the flow of gas from inner zone into the fracture; the cylindrical
shape with blue outline represents the well; the cuboid represents the reservoir inner zone;

Figure 3—Schematic of a flow from reservoir into the fracture; outer rectangle represents the boundaries
of the reservoir (arbitrarily chosen); the inner dashed rectangle represents the inner zone or the
stimulated reservoir volume; the Tri-linear model is represented in the above schematic using the red
arrows which show the direction of gas flow into the inner zone, fracture, and eventually into the well;

Due to the low permeability of the rock surrounding the fracture, the flow from stimulated rock to the
fracture is expected to be dominated by "linear" flow. "Linear" flow is a term used to represent flow along
a single dimension with the assumption that the pressure gradients are mainly in that direction. Thus a
complex 3-dimensional flow problem is broken down into a series of 1-dimensional flow problem which
is much easier to solve analytically with the appropriate boundary conditions. Similarly the flow from the
fracture to the wellbore is expected to be a "linear" flow regime. The coupling of the regimes is due to the
boundary condition of inflow from rock to fracture.
SPE-195798-MS 5

Figure 4—Schematic describing the flow of gas condensate or volatile oil flow in the reservoir and in fracture
The condensate banking in the rock and also in the fracture can cause reduction of gas well deliverability.

The quantification of the fracture surface area created during fracturing is a challenge. Indirect methods
such as matching the pressure time or rate time behavior of the well production may be used to infer the area
created. In additional, history matching a reservoir simulation model may be used to quantify the reservoir
fracture characteristics.
History matched models lend validity or increase the confidence in our understanding of the reservoir/
fracture characteristics. Thus the parameters used in the model may be compared with other independent
source of reservoir/fracture diagnostics.
As described above the Trilinear Model (from Brown et al., 2011) is used to simulate the well
performance. In the following section a regression model is defined to understand well performance.
However, the regression modeling is an empirical method to understand the behavior of the wells but can
be used to identify cause and effect relationships.
In order to obtain a physics based understand we construct reservoir models and introduce flow behavior
of multiple phases in the reservoir, fracture and the wellbore. The workflow followed in constructing a
representative simulation model is as follows.
1. Assume fractured reservoir parameters and simulate the historical rates of gas and condensate;
2. If the historical rates match (visual inspection) then the model parameters assumed are reasonable
and may be used for forecasting;
3. If the match is not good (for instance high gas rates towards end of history) then consider revising
the model parameters as follows;
4. Check the fracture length and height assumed;
5. Check the inner zone permeability assumed – usually this is subject to interpretation as the inner zone
represents the stimulated reservoir volume which is modeled with an effective permeability which
cannot be directly measured;
6. Check the stress dependent fracture permeability parameters and use laboratory data as much as
feasible;
7. Correlate with other estimates of fracture length, height and width – such as microseismic or other frac
modeling software (note that the trilinear model already incorporates an inner zone that represents
the stimulated zone).

Simulation of Well History


History Match of Well A. The well performance of Well A is simulated using the workflow used above.
Assumptions of model are:
1. No adsorption used;
2. No change in fracture conductivity with pressure or time for the first simulation; the second simulation
considers stress dependent permeability;
3. PVT behavior simplified with condensate gas ratio value at dew point – this can lead to incorrect
condensate volumes in the near fracture region.
The impact of stress dependent permeability appears to be significant. The estimated inner zone
permeability is a representation of both the fracture conductivity in the stimulated reservoir volume and the
effective surface area created due to the fracture operation.
6 SPE-195798-MS

Figure 5—Gas-water and gas-oil relative permeability curves along with the limiting saturations.

Figure 6—Oil-water relative permeability curves used in the model simulation.

Table 1—Properties of the fractured reservoir that best fit the production data history of Well A.
SPE-195798-MS 7

Table 2—Assumptions and known parameters used in modeling the production well performance history of Well B.

Phase Behavior Description. In order to describe the phase behavior a modified black oil type model is
used. The black oil model is a well established modeling technique where the properties of the crude oil
are represented using a 2 component system and 2 phases. The three components and phases are liquid and
gas. The change in volume of gas with pressure is represented using the formation volume factor which is
nothing but a ratio of the reservoir volume to that of the surface volume. In the process of transporting the
fluid from subsurface reservoir to the surface the phases separate and in the case of oil, the volume shrinks.
The Pressure Volume Temperature (PVT) thermodynamic properties of the gas phase is described using
the Benedict Webb-Rubin model. In order to adapt the method to the gas condensate systems which are
usually retrograde gas condensates, two additional terms are defined. The first is the "vaporized oil ratio"
term "Rv" which is nothing but the volume of hydrocarbon that will condense into liquid per unit volume
of separator gas that is produced. The value for Rv is a function of pressure at isothermal conditions. A
rigorous calculation of the Rv requires solution of an equation of state for a given composition of the fluid.
A constant Rv indicates a constant value for the quality (liquid to gas volume ratio) on the P-T diagram for
the condensate. Ideally an EOS model can calculate at each pressure the quality and composition of the two
phases. However, it is more convenient to use empirical correlations such as the Ovalle et al., 2007.
The second term that is introduced to modify the conventional black oil model to adapt for a condensate
system is the dry gas formation volume factor (Bgd). The dry gas formation volume factor is a ratio of the
volume of the gas at the given reservoir condition to that at the standard conditions. The difference is that
the dry gas formation volume factor does not consider the volume of condensate dropped out of the gas
phase after the pressure reduction at surface. The model formulated by Whitson and Brule, 2000 is used to
characterize the formation factor of the dray gas.
8 SPE-195798-MS

Figure 7—Vaporized oil ratio (Rv) used in the empiricial modeling of gas condensate
behavior of the liquids rich gas encourntered in the Willesden Green wells.

Model Prediction.

Figure 8—Plot of production of gas (red data points and solid line) and condensate (green data points and solid line)
from Well A. Model match with the data is shown in the solid lines. The brown data points and solid lines show the
pressure control used in the simulation. Notice the poor match in the absence of stress dependent permeability.
SPE-195798-MS 9

Figure 9—Ratio of permeability at a given pressure that at initial or reservoir pressure conditions. As
reservoir pressure decreases the ratio decreases indicating the reduction in the permeability with depletion.

Figure 10—Ratio of formation compressibility with respect to initial formation


compressibility at reservoir conditions of pressure. As with permeability the formation
compressibility undergoes change to lower values with decrease in reservoir pressure.

Figure 11—Plot of production of gas (red data points and solid line) and condensate (green data points and solid line)
from Well A. Model match with the data is shown in the solid lines. The brown data points and solid lines show the
pressure control used in the simulation. The above plot is constructed with stress dependent permeability included.
10 SPE-195798-MS

Figure 12—Plot of normalized pressure drop versus square root of producing time for Well A. The end of linear flow is
indicated by the vertical bar and the data points prior to the end of linear is indicative of the linear flow regime from the
matrix into the fracture. The slope of this line is a product of the fracture half-length and the square root of permeability.

A common technique to determine production deliverability in unconventional shale plays is to perform


analysis of the transient time period using linear square root time plots. The linear square root time plot
creates a straight line through the production data to qualitatively determine production deliverability with
the term Xf√k. The Xf√k relates the total area for flow with an estimate of system permeability.
The estimated fracture width based on the dimensionless expression in Brown et al., 2011, model, comes
out to be ~0.08 feet (~2.4 mm). The fracture width appears to within reasonable limits although slightly
on the higher side. The fracture length used in the model is 160ft which is also reasonable and within the
estimated region of the microseismic event cloud.
History Match of Well B. Using a similar workflow to that of Well A, a second well, Well B, was history
matched to estimate the parameters that describe the stimulated reservoir. Well B is a poorly performing
well among the set of wells considered in this study with gas and oil volumes about 1/4th that of the Well A.
SPE-195798-MS 11

Figure 13—Plot of production of gas (red data points and solid line) and condensate (green data points and solid line)
from Well B. Model match with the data is shown in the solid lines. The brown data points and solid lines show the
pressure control used in the simulation. The above plot is constructed with stress dependent permeability included.

Figure 14—Plot of normalized pressure drop versus square root of producing time for Well B. The end of linear flow is
indicated by the vertical bar and the data points prior to the end of linear is indicative of the linear flow regime from the
matrix into the fracture. The slope of this line is a product of the fracture half-length and the square root of permeability.
12 SPE-195798-MS

Table 3—Properties of the fractured reservoir that best fit the production data history of Well B.

Table 4—Assumptions and known parameters used in modeling the production well performance history of Well B.

In Group 1 wells, fractures in the SRV may be either closing or loaded with condensate or both, which
requires further analysis. The model without stress dependent permeability predicts higher production than
observed.
It appears that, both in the Well A and Well B, the reservoir and the fracture properties change with
pressure and therefore with time. The introduction of the stress dependent permeability accounts for a
portion of these changes. Condensate loading may also increase as pressure decreases. One hypothesis is
that the fractures close over time and hence the fracture conductivity and hence the Xf√k value decreases
with time. This hypothesis finds credence in the RTA analysis and hence can very well be true.
In the above case, it will advantageous to use large proppant volumes to increase the initial fracture
conductivity and hence mitigate the decrease in the well performance and result in higher average
cumulative 6 month production volumes.
The estimated effective permeability for the wells show a similar trend with the Xf√k analysis derived
from the RTA. The well with higher estimated effective inner zone permeability also has a similarly higher
Xf√k value. Figure 15 shows the trend for a total of nine wells whose production data were matched with
a model. More number of wells may be studied in a similar fashion to confirm the behavior.
SPE-195798-MS 13

Figure 15—Plot of estimated Xf#k versus estimated fracture conductivity from the trilinear model history match runs.

Regression Analysis Using Multivariate Models and Classification Methods


Multivariate Regression Method
A multivariate regression of production and completions data collected for Group 1 and the entire region,
inclusive of the Group 1, shows consistent trends. The main response variables considered are the produced
volume averaged over a time period and the expected ultimate recovery (EUR). EUR data were considered
only for the 15 well dataset for Group 1 region. The geologic parameters considered were:
1. Latitude (~330 Wells from Duvernay)
2. longitude (~330 Wells from Duvernay) and the
3. true vertical depth of the well (TVD depth) (~330 Wells from Duvernay).
The operational parameters considered were
1. Total perforation clusters (Encana's 15 well set)
2. Completed length of lateral (~330 Wells from Duvernay)
3. Number of stages (~330 Wells from Duvernay)
4. Proppant weight used (~330 Wells from Duvernay)
5. Liquid volumes used (~330 Wells from Duvernay)
6. Average proppant weight used per length (tons), (~330 Wells from Duvernay)
7. Average liquid volume used per length (m3), (~330 Wells from Duvernay)
8. Wellbore direction
9. Condensate to Gas Ratio (CGR) calculated based on 6 month cumulative volume produced
10. Type of stimulation fluid (1-Slickwater; 0.5 Hybrid; 0-Gelled)
11. Hybrid Frac? (1-Yes; 0-No)
Some preliminary conclusions from that analysis are discussed below.
A recurring issue with linear regression is that the trends do not indicate the optimal conditions of the
above parameters but only local trends. This limitation is inherent to linear regression and hence more
sophisticated techniques are required to understand the optimality conditions. Two factors, indicating the
fluid type are used, to identify the impact of fluid type.
14 SPE-195798-MS

Because of the above reason, the results of regression analysis should be understood with the relevant
physics of the problem. For instance, a positive regression coefficient for the parameter TVD would seem
to suggest that the most optimal is 6 month cumulative production is achieved with deeper wells. However,
we clearly know from geology that the depth is constrained by the interval thickness, which in this case is
limited by the upper and lower Duvernay.
Regression modeling of the region shows that the wells with better performance are in the western
periphery and towards the south with greater depths. The performance metric considered is the 6 month
cumulative production in barrels of oil equivalent. This may be concluded from the sign of the regression
coefficients for latitude (negative), longitude (negative), and the TVD (positive). The main factors that show
impact (as determined by the magnitude of the absolute value of the coefficient and the low p-values) are:
1. Total proppant placed per well (tonnes); 2. Condensate fraction (defined as CGR/[1+CGR], where CGR
is the condensate to gas ratio based on 6 months cumulative produced volume);
See Table 3 for details on the regression analysis along with the relevant analysis of variance. The 520
dataset was reduced to 330 wells when the wells with zero 6 month cumulative production were removed
from the analysis. The indicator for goodness of fit is the R2 value and the significance of the individual
regression coefficients are given by the t-test and p-value calculations provided in the table.

Table 3—Summary of Greater Duvernay Regression Modeling. All the variables are centered and normalized with the
standard deviation. The highlighted factors show a greater impact on the 6 month cumulative volume of fluids produced.

Those factors that showed poor p-values were removed from the model, leaving the remaining significant
factors. Thus, not all the factors listed in the previous section are observed in Table 3 because they were
deemed insignificant without loss in goodness of fit.
SPE-195798-MS 15

The regression model for Greater Duvernay shows that large amounts of proppant used leads to observed
improved well performance. This invariably also leads to large volumes of fluid which is indicated by the
positive coefficients for both. The coefficient for proppant weight used is twice that of the liquid volumes
used. However, a negative coefficient for the average fluid pumped per meter indicates that the fluid volumes
used were less per stage for improved performance. However, the regression coefficient is relatively smaller
compared to the proppant volume used.
The observation that more proppant weight per well and less fluid volume per unit length of the well
leads to improved performance, shows that this may be an outcome of the large number of hybrid and
crosslinked gel type fluids used in the trials in Duvernay overall. Unfortunately the dataset has a poor clarity
on the exact type of fluid used and therefore, an explicit regression on this variable was not possible in this
exercise. Many of the wells are indicated to be a hybrid fracture or slickwater, but the fracturing fluid used, as
indicated in the database, is really a cross-linked fluid. Also the sequence of fluids used in different times of
fracturing in a stage cannot be adequately represented in a single factor. This issue may also be affecting the
interpretation of the regression coefficient pertaining to the proppant concentration which shows a slightly
negative value.
The condensate fraction, a measure of the fraction of the condensate shows a strong impact on the 6
month cumulative production. It must be noted that the fraction was calculated with the CGR represented in
units of Barrels per million standard cubic feet of gas produced. Thus the denominator in the fraction really
considers a reference CGR of 1 Barrel/MMscf. The distribution of the variable is less skewed. It is clear that
the lower condensate fraction along with greater amounts of proppant per well correlates with the observed
6 month cumulative volume produced. The higher proppant volumes, placed per well, generally also leads
to improved fracture conductivity, which in turn leads to better flow rates of gas and condensate unloading.
Figure 16 shows the comparison of predicted 6 month cumulative production (using the regression model)
and the actual measured values. The R2 value is somewhat low. One reason for this could be the non-linear
behavior of variables which are not possible to account for in the linear regression method.

Figure 16—Plot of the predicted and measured values of the 6 month cumulative production
in Greater Duvernay. The values are centered and normalized with the standard deviation.

In Group 1 region, the regression analysis shows that higher fluid volumes per meter (positive coefficient)
combined with lower proppant weights per meter (negative coefficient) correlated with observed well
performance. See Table 4 for a presentation of the analysis. Additionally, lower total fluid pumped volumes
16 SPE-195798-MS

seem to correlate better with 6 month cumulative production which may be due to efforts to contain fracture
growing out of the zone of target. Thus small treatment jobs with lower volumes and lower proppant weights
per well correlate with the observed well performance in Group 1 region.

Table 4—Summary of Group 1 region Regression Modeling. All the variables are centered and normalized with the standard deviation.

However, it must be noted that the majority of the well treatments in the Group 1 region are slickwater
treatments. The slickwater treatments inherently place lower proppant weights for a given fluid volume.
Also the data seems to indicate that the treatment volumes are lower, which may be a result of fracture
containment concern in Group 1 forcing the operators to use smaller volumes of both proppant and fracturing
fluid volumes.
In contrast to the greater Duvernay region, condensate fraction does not significantly impact the produced
volumes. The coefficient is relatively smaller (half as much as that in the greater Duvernay) and in fact the
p-value is much greater than 0.05 for the 95 % confidence limit indicating that the alternative hypothesis,
that the condensate fraction does not affect the response, may be true. Thus, there may be other factors, not
explicitly accounted for in the regression model that affects the 6 month produced volumes.
As before, we don't have a clear dataset which identifies the type of fluid used in the wells and so it is
not possible to identify the behaviors related to gelled fluids separately. In any case there seems to be less
gelled fracturing in Group 1. This may be a significant factor in considering the future operations in Group
1 to improve production.
SPE-195798-MS 17

The Greater Duvernay wells perform better with lower fluid volumes and higher proppant weights. The
Group 1 region well performance also correlates with lower fluid volumes. As discussed earlier, the lower
fluid volumes, per well, used in Group 1 region may be due to concerns related to fracture containment.
This however leads to lower proppant placement also, especially when using slickwater fluids as in Group
1 region. Most of the trials in the Group 1 are slickwater treatments, which inherently allow only lower
proppant weights. The average Group 1 region 6 month production (447 MBOE) is lower than that of the
Greater Duvernay (631 MBOE) likely because of this reason.
The models for both Group 2 region and the greater Duvernay region show a strong impact of the total
proppant placed on the 6 month produced volumes. Group 1 model on the other hand does not show a clear
dependence – in fact the total proppant placed appears to correlate negatively with the 6 month produced
volume.
Therefore, a potential solution to improving the 6 month cumulative production in Group 1 region may
be to consider methods to place larger proppant volume using a viscosified fluid treatment with lower fluid
volumes and higher proppant weights. Essentially this is to pack more proppants with lower fluids. Other
variations of this method may be attempted to reduce costs.
The regression model prediction of 6 month cumulative production is compared (Figure 14) with the
actual measurements which show a reasonably good R2 value. However, more data is needed to predict the
trend if the fluid type is changed. At present, most of the well fracturing treatments are based on slickwater.

Figure 17—Plot of the predicted and measured values of the 6 month cumulative production
in Group 1 region. The values are centered and normalized with the standard deviation.

A separate regression model analysis with just the Group 2 region wells showed a strong dependency
on the total proppant placed (Table 5). It must be noted that this is consistent with the model estimated for
the Greater Duvernay, but contrasts with the Group 1 region model. As in the Greater Duvernay model,
the condensate fraction in the Group 2 region also has a strong impact on the observed 6 month produced
volumes. This again shows that the higher performing wells in the region requires placement of large
amounts of proppant.
18 SPE-195798-MS

Table 5—Summary of Group 2 region Regression Modeling. All the variables are centered and normalized with the standard deviation.

The condensate fraction may not be a causal factor but a correlated factor with other parameters
considered. The increase in proppant amount placed, along with the type of fluid may impact the ability
of the well to unload condensate and hence result in better well performance. This is apparent from the
coefficient for the average fluid pumped per meter, which is much smaller than that in the Group 1, indicating
that less fluid volumes per meter have been used for the wells in Group 2 region compared to the Group 1.

Figure 18—Plot of the predicted and measured values of the 6 month cumulative production
in Group 2 Area. The values are centered and normalized with the standard deviation.
SPE-195798-MS 19

A plot of the proppant placed per well, in tonnes, with the 6 month cumulative produced volume for Group
1 and the Group 2 regions is shown in Figure 16. It is clear from the plot that the total proppant amount used
in the Group 2 region is higher and leads to greater 6 month volumes produced. However, there appears to
be a significant scatter at a given amount of proppant placed. This could be due to the different proppant
placement strategies adopted in the regions. Further study is required to understand the model better and
identify other parameters that affect the performance. For instance, the bottomhole sandface pressure is a
variable which can affect the well performance. Currently, this data is not available for all the wells in the
regions considered.

Figure 19—Comparison of the impact of total proppant placed per well on


the 6 month produced volumes in the Group 2 area and the Group 1 area.

Random Forests Classification Method

Model Without Condensate Effects

Figure 20—Random Forests Regression Results for the traiing set –Without Considering Condensate
Production Effects on the Well performance. The R2 value for the first plot on the left is 0.95, the R2
value for the 6 month production fitting result is 0.93 while that for the 12 month production set is 0.90.
20 SPE-195798-MS

Figure 21—Random Forests Regression Results for the testing set –Without Considering Condensate
Production Effects on the Well performance. The R2 value for the first plot on the left is 0.65, the R2
value for the 6 month production fitting result is 0.74 while that for the 12 month production set is 0.62.

Table 6—Importance coefficients for predicting initial production, 6 month cumulative


production, and 12 month cumulative production respectively from left to right.

Model Including Condensate Production

Figure 22—Random Forests Regression Results for the traiing set –With Considering Condensate
Production Effects on the Well performance. The R2 value for the first plot on the left is 0.89, the R2
value for the 6 month production fitting result is 0.92 while that for the 12 month production set is 0.89.
SPE-195798-MS 21

Figure 23—Random Forests Regression Results for the testing set –Without Considering Condensate
Production Effects on the Well performance. The R2 value for the first plot on the left is 0.70, the R2
value for the 6 month production fitting result is 0.72 while that for the 12 month production set is 0.75.

Table 7—Importance coefficients for predicting initial production, 6 month cumulative


production, and 12 month cumulative production respectively from left to right.

Conclusions
1. Group 1 well models show the potential for much higher production than observed in the absence
of stress dependent permeability. The wells studied in article are affected by stress dependent
permeability, fracture closure, or condensate loading.
a. Model predictions shows that the reservoir is capable of producing higher volumes of gas and oil
if the stress dependent effects on permeability and SRV fracture conductivity can be mitigated;
b. All of the above may be mitigated by following a strategy of higher proppant usage with limited
fluid used to increase the fracture conductivity;
2. Greater Duvernay region well performance is on average higher than the Group 1. Based on
multivariate regression methods, the following inferences may be drawn.
a. Higher proppant concentrations are used, per well, with limited fluid volumes correlates with
better well performance;
b. The converse seems to be true for Group 1.
3. Based on the Random Forests Classification technique and Multivariate Regression Modeling
technique it can be said that total proppant placed contributes more to production in short period;
Creating longer fractures and increasing fracture density contributes more to production in the longer
term.
4. Gelled fluid fracturing may be explored further in Group 1 region to see if the well performance can
be improved even further.
22 SPE-195798-MS

a. Use of gelled fluids can limit fluid volume and increase proppant concentration;

References
Benedict, M., Webb, G. B., and Rubin, L. C. 1940. An Empirical Equation for Thermodynamic Properties of Light
Hydrocarbons and Their Mixtures I. Methane, Ethane, Propane and N-Butane. The Journal of Chemical Physics 8(4):
334.
Brown, M., Ozkan, E., Raghavan, R., & Kazemi, H. (2011, December 1). Practical Solutions for Pressure-Transient
Responses of Fractured Horizontal Wells in Unconventional Shale Reservoirs. Society of Petroleum Engineers.
doi:10.2118/125043-PA.
Carr, N. L., Kobayashi, R., & Burrows, D. B. (1954, October 1). Viscosity of Hydrocarbon Gases Under Pressure. Society
of Petroleum Engineers. doi:10.2118/297-G.
Clarkson, C. R., & Williams-Kovacs, J. (2013, July 4). Modeling Two-Phase Flowback of Multifractured Horizontal Wells
Completed in Shale. Society of Petroleum Engineers. doi:10.2118/162593-PA.
Clarkson, C. R., Qanbari, F.,Nobakht, M., & Heffner, L. (2013, August 1). Incorporating Geomechanical and Dynamic
Hydraulic-Fracture-Property Changes Into Rate-Transient Analysis: Example From the Haynesville Shale. Society of
Petroleum Engineers. doi:10.2118/162526-PA.
Dobrynin, V. M. (1962, December 1). Effect of Overburden Pressure on Some Properties Of Sandstones. Society of
Petroleum Engineers. doi:10.2118/461-PA.
Gobran, B. D.,Brigham, W. E., & Ramey, H. J. (1987, March 1). Absolute Permeability as a Function of Confining
Pressure, Pore Pressure, and Temperature. Society of Petroleum Engineers. doi:10.2118/10156-PA.
Ovalle, A. P., Lenn, C. P., & McCain, W. D. (2007, December 1). Tools To Manage Gas/Condensate Reservoirs;
Novel Fluid-Property Correlations on the Basis of Commonly Available Field Data. Society of Petroleum Engineers.
doi:10.2118/112977-PA.
Ramey, H. J. (1964, April 1). Rapid Methods for Estimating Reservoir Compressibilities. Society of Petroleum Engineers.
doi:10.2118/772-PA.
Vyas, A., Datta-Gupta, A., & Mishra, S. (2017, November 13). Modeling Early Time Rate Decline in Unconventional
Reservoirs Using Machine Learning Techniques. Society of Petroleum Engineers. doi:10.2118/188231-MS.
Williams-Kovacs, J. D., & Clarkson, C. R. (2013, November 5). Stochastic Modeling of Multi-Phase Flowback From
Multi-Fractured Horizontal Tight Oil Wells. Society of Petroleum Engineers. doi:10.2118/167232-MS.

Potrebbero piacerti anche