Exploration Insights December 2019/ January 2020 | Page 10

Exploration Insights | 11
Application of a Machine
Learning Approach for
Predicting and Classifying
Production Success in
Unconventional Plays
by: Jessica Wevill, Imperial College, U.K.
A road cut of the Eagle Ford shale from Val Verde County, Texas, which is on the border with Mexico.
LOOKING FOR A MASTER’S PROJECT?
Apply for the STEPS program - the tender window opens in November.
The STEPS program provides data, software, training, and mentorship to post-
graduate master’s students completing their independent research thesis. This
year, our STEPS projects will focus on “Planet to Pore Modeling.”
As part of the STEPS outreach program, students will have the opportunity to:
• Experience real-world data
• Utilize large and disparate datasets
• Define a methodology to resolve an issue or make an interpretation
• Work within the cloud environment
• Gain experience with industry-standard computer applications
Visit our website for more information.
Signaling the start of the “Golden Age of Gas”
(McGlade et al., 2013), the introduction of
economically viable hydrocarbon production from
tight shale reservoirs has changed the future of
the U.S. natural gas industry, prompting a global
shift in the petroleum industry towards the
exploration and production of shale reservoirs.
The North American shale revolution can,
therefore, be used to help inform successful
exploration and efficient production from
unconventional resource plays, globally. WHAT IS MACHINE LEARNING?
The complexity of fracture networks and
geological heterogeneities means that prediction
of production, even between two adjacent
wells within the same play, is challenging. The
majority of work regarding unconventional plays
has been conducted on a play-by-play basis in
order to understand internal heterogeneities,
and to improve production and well placement.
Geological factors influencing hydrocarbon
production are interlinked and cannot be treated
independently. This article demonstrates how a
machine learning (ML) approach to classifying
unconventional plays can reveal underlying trends
between geological factors and production
success across seven U.S. plays, defined in
Figure 1. THE DATASET
Advances made in artificial intelligence (AI) as a
method of processing big data are increasingly
used within the geosciences and in the petroleum
industry to improve the efficiency of tasks and
accuracy of the predictions. ML provides a means
of harnessing a vast supply of data, and the
understanding of internal data structure, through
data mining. ML is inherently more accurate than
learning by humans, and avoids human bias.
The dataset in this study comprised subsurface
property data, production data, and completion
metrics for >135,000 lateral wells. The
subsurface property data were derived from basin
scale geological models generated by Neftex ®
Insights, and were attributed to each lateral based
on the intersection between the surface well
coordinates and the property grid. Production and
completion metric data were provided by Rystad
Energy (ShaleWellCube). This vast volume of data
points formed the input data for algorithms to
learn from and predict production metrics.