American Oil and Gas Reporter - July 2016 - 77

SpecialReport: Big Data & Geostatistics
end in analysis paralysis," she says.
There also are patterns in human behavior that must be accounted for, says Ball. "Humans operate differently from machines and
from one another," she observes. "The same equipment in the same
environment interacting with different people can have different
results. The analysis must be informed by human behavior."
All data are replicated and stored in a "reservoir" that connects
disparate data from disparate systems. "The data reservoir represents the truth for us. It is refreshing constantly so that data are
always up to date," she explains, adding that Devon also has some
vendor data in cloud and hybrid cloud configurations. "This level of data integration makes data readily available to support decision making. Users can slice and dice data any way they want."
Analytics Infrastructure
Before embarking on an analytics project, Ball says Devon's
team considers the four "V" characteristics of the associated data:
volume, variety, velocity and validity. The largest sets of data Ball's
team has encountered were terabytes in size, occasionally approaching an exabyte, but she says datasets normally are sized
in gigabytes.
Using fit-for-purpose infrastructure and open-source technologies allows the team to cost effectively analyze the full dataset, storing data and running applications on commodity hardware clusters.
"We used to have to look at data samples because of size limitations," Ball recalls. "With the new framework, we can look at
entire basins with no trouble. We pull in the data, create an analysis, and output the results in a form that can be consumed easily."
One of the keys to Devon's success with data analytics has been
the support from its information technology group, Ball says. "My
sense is that there generally is not a high level of comfort in IT
departments with using open-source technologies, but in our case,
we are collaborating-first and foremost-on the mission to find
and produce oil and natural gas."
The variety of data the team works with continues to broaden, and includes a range of geological and geophysical data (including seismic, petrophysics, geology and reservoir measurements), completions and production data, and business and financial data. The data come from different sources and in different
formats, from numerical and text to images and video.
"We analyze text, for example, for potential improvements in
nonproductive time," she illustrates. "Or we may analyze images
and associate them to temperature and pressure changes as part
of a predictive asset failure program.
"Our challenge is to tie all these data together and run an analysis that produces meaningful, strategic and actionable results,"
she continues. "We use a range of analytical techniques, including clustering analysis and supervised learning. The choice of technique depends largely on what is known about the data."
Velocity has two considerations: How fast data are needed, and
how fast decisions must be made. One of the key questions to endusers is what real time means to them. "We get answers ranging
from once a week to once every second," replies Ball. "It is im-

portant to define what you really need and really can act on."
In critical dynamic operations such as drilling or fracturing in
which real-time decisions are needed, in-memory and in-database
technologies are preferred. "This takes the analysis to the data, rather
than dragging the data to an external tool," she explains.
Validity refers primarily to data quality control, particularly
the ability to screen noise and bad data, Ball reveals. "With machine learning and artificial intelligence, if the model does not
know there are bad data, then the model is going to deteriorate,"
she reasons. "You have to be able to test data quality in real time
to determine whether it is within the bounds of reasonable."
Learning New Languages
Devon's cross-functional analytics team consists of members
with specialization in data science and embedded IT, as well as
representatives from oil and gas disciplines. "I have a drilling engineer and a reservoir engineer being trained as data scientists,"
she remarks. "These folks are bridges between us and the subject matter experts in Devon's various departments."
Analytics requires learning many languages, according to Ball.
"And not all of them are computer languages!" she quips. "The
biggest surprise to me was how rarely a data scientist spoke the
same language as a geoscientist or engineer. A lot of translation
is required. Geophysicists and reservoir engineers go straight to
the applied mathematics. They see everything through the discipline of their science, so we have to meet them where they are.
A good data scientist is insatiably curious and wants to know why.
We strive to hear the story the data want to tell, not what we expect to hear."
Rather than analyzing data in a more natural dynamic state,
oil and gas companies traditionally have examined them "at rest,"
Ball points out. "The data were captured at a given point in time,
entered into a storage system, processed, and then the interpreter
looked for an answer," she says. "The process may have taken a
day, a month or a year or longer."
In contrast, streaming in-memory analysis allows almost instant answers. "End-users are increasingly able to see real-time
analysis in three dimensions in a collaborative space, and even
in motion," Ball concludes. "The analysis itself and the results
are reflecting the reality of the operations from which data are
collected, in that they are becoming increasingly dynamic and
moving in real time."
Ball points out that data science is regarded by many experts
as the most critical skill set required across all industries going
forward. In oil and gas, specifically, the business case for analytics is powerful already. Looking toward the future, Ball assess
the defining role technologies such as analytics, data mining, big
data and self-learning machines ultimately will have on the geosciences and petroleum engineering.
"It is exciting to see the next generation embrace analytics and
data science as a complement to physics-based engineering," she
enthuses. "Just imagine the leaps our industry can make with that
powerful combination!"
r
JULY 2016 77

American Oil and Gas Reporter - July 2016

Table of Contents for the Digital Edition of American Oil and Gas Reporter - July 2016