American Oil and Gas Reporter - August 2017 - 77

SpecialReport: Horizontal Wellbore Construction
Once the analyses have been completed
and the slopes have been calculated, the
impact of all parameters on the production
index can be determined and plotted in
the form of a tornado chart of KPIs.
Figure 4 shows the impact of different
parameters on 30-day cumulative production. Grouping these parameters and
averaging their impact shows that natural
and completion design parameters have
very similar impacts on the production
index in this particular Marcellus asset.
Furthermore, although not readily apparent
in this image, the analysis shows that
this similarity does not change with time.

Predictive Modeling
Four types of models are used for shale
wells: decline curve analysis (DCA), rate
transient analysis (RTA), numerical simulation and data-driven analytics. It is important to be cognizant of the fact that all
of these shale modeling techniques include
assumptions. Being aware of the assumptions involved in a given technique is the
most important part of developing and
working with models derived from it.
DCA models are the simplest, essentially representing a statistical curve fit
of production data. No parameters other
than production rate are used in DCA,
and this simplicity makes DCA an attractive tool. However, the assumptions made
when using DCA include boundary-dominate flow, single-phase production, homogeneous reservoir characteristics, constant bottom-hole pressure and no changes
in operational constraints throughout the
life of a well. Likewise, there are numerous
assumptions made in developing RTAbased and numerical simulation models.
Assumptions involved with developing
data-driven models in shale analytics
mainly concern the data that are used to
develop the model. They can be summarized as assuming the data are sufficient
in quality and quantity to develop a predictive model; include the necessary information for decision making; and are
representative of well construction, reservoir, completion and production.
The steps in developing a data-driven
predictive model include:
* Selecting the input parameters.
Input parameters are usually independent
of one another, and the number of wells
in the analysis dictates the number of
input parameters that can be used in the
model. It is possible, and sometimes necessary, to use dependent variables as
input to the data-driven predictive model.
However, this only should be done if the

tool being used for predictive model development has the capability to incorporate
dependencies between several parameters
in the models, and keep track of and
honor all dependencies throughout the
analyses (not a simple task).
* Randomly partitioning the data into
three segments. One segment is used to
train the model, one is used to oversee
the training process to avoid overfitting
and memorization (it is not accessible to

the data-driven model), and one is excluded
from the training and calibration processes
and used as blind validation data.
* Selecting the technology for developing the model (it must be supervised,
and should not be rule-based to minimize
bias in the system).
* Selecting the learning algorithm.
* Training.
* Validating predictions using the
blind well data.
r

The Ultimate Diverter™

™

Old fractures isolated

New fractures through ESeal RF Liner

How can you permanently seal off older,
unproductive perforations?
Particulates or chemical diverting agents are unpredictable, unreliable, and
often unsustainable. Enventure's ESeal™ RF (Refrac) Liner is a better
answer. It reliably creates a new wellbore with permanent isolation of existing
perforations and internal pressure integrity. In practice, this results in a
faster payback on investment and extended production life of the reservoir.
Enventure's ESeal RF Liner offers a single, permanent process, maintains
pressure integrity, and provides more accurate, more predictable diversion.
It's cost-effective, reliable, and permanent. In other words, it's the
Ultimate Diverter.
To find out more, visit us at: www.EnventureGT.com/refrac

AUGUST 2017 77

http://www.EnventureGT.com/refrac

American Oil and Gas Reporter - August 2017

Table of Contents for the Digital Edition of American Oil and Gas Reporter - August 2017