Page 158 - Intelligent Digital Oil And Gas Fields

Components of Artificial Intelligence and Data Analytics 121

It is interesting to note that in the applications of advanced data analytics, a
small number of methods always seem to perform better than most others.
This phenomenon has been documented for classification problems
(Fernandez-Delgado et al., 2014) in a study that combined 179 classifiers
from 17 families, while using 121 data sets. In petroleum engineering and
geoscience applications that largely pertain to DOF architecture as well,
many of Big Data problems involve regression. However, it seems like
the same set of methods always outperform others: (artificial) neural net-
work, support vector machine (SVM), and random forest. Depending on
the nature of the application, some of these methods are more suitable than
others, but only rarely does the user need to look outside of this suite to more
exotic methods. Section 4.3 briefly captures some of the prominent appli-
cations of data analytics from the E&P domain, pertaining to DOF projects.
Although the field of ML is a relatively young one, it is rapidly growing.
One of the emerging trends in ML is the so-called recommendation systems
(Leskovec et al., 2014; Jordan and Mitchell, 2015), applications that involve
predicting user responses to different options. One genre of recommenda-
tion systems is the area of collaborative filtering, which recommends items or
actions based on similarity measures between the users and/or items. Such a
paradigm naturally appeals to the concept of DOF, which is by definition a
collaborative environment [e.g., real-time operation center (RTOC)],
where oilfield operators work with advanced sensor- and data-driven
technology.
Maybe the future of ML techniques for DOF and data-driven E&P oper-
ations is hidden in the analogy to natural learning systems. As envisioned by
Jordan and Mitchell (2015), this concept suggests the idea of team-based,
mixed-initiative learning. In a nutshell, since the current ML systems mostly
operate in isolation to analyze given data, people often work in teams to col-
lect and analyze data, by bringing together a variety of expertise and perspec-
tives when solving particularly complex and difficult problems (e.g., large-
scale integrated reservoir studies). Perhaps in the next-generation DOF, new
ML methods will work collaboratively with oil and gas field operators to
extract and mine deep knowledge and subtle statistical regularities from mas-
sive data sets (Big Data) acquired at extreme velocities and frequencies by
smart IoT sensors, to generate intelligent real-time operational decisions.
For completeness and to facilitate easier understanding of the applications
presented in Section 4.3, we present a quick overview of the mentioned ML
techniques.