Page 147 - Intelligent Digital Oil And Gas Fields

110 Intelligent Digital Oil and Gas Fields

correlations, mostly of a highly nonlinear nature (see Table 4.3 for a clas-
sification of the main tools and techniques gathered under the umbrella
of AIPA). Examples of predictive analytic implementations include
modeling, equipment operations, production trends, reservoir dyna-
mics, and asset failure predictions to minimize downtime. Financial
predictive algorithms compute expected response and ROI for work-
overs, pump changes, injection rate modifications, drilling plans, and
alternative completion parameters. However, recent surveys indicate
that less than1% of companies surveyed have actually deployed predic-
tive analytics. The ones who have found extremely encouraging results
that have added significant values to their businesses.
• Prescriptive analytics apply the outcomes and insights of predictive analyt-
ics and turn them into actionable foresights by applying advanced process
optimization methods. Accurate predictions help us understand the
actions to be taken to maximize good outcomes and minimize or prevent
potentially bad outcomes. Examples of prescriptive analytics include
alerts (e.g., opportunities, abnormalities, and data problems), recom-
mendations (e.g., the workover to perform next, when to stimulate a
well, optimal injection rates throughout a waterflood, or where to drill),
and optimization (e.g., capital allocation, investment, and risk manage-
ment). Currently, R&D in predictive analytics for E&P is cutting edge;
for example, a recent breakthrough includes a novel data—physics par-
adigm in modeling and optimization of oil and gas assets (Sarma and
Leport, 2016).
To conclude this section, we propose a schematic example of a modular
advanced data analytics workflow for the E&P industry, where components
of individual analytics domains, from descriptive to prescriptive, merge into
a collaborative synergy (Fig. 4.3). The workflow consists of five modules and
begins with the database management module, which includes data acqui-
sition (i.e., data from smart sensors/IoT, etc.), integration [i.e., subsurface
(geological and geophysical), drilling and completions, production, stimula-
tion, operations], and aggregation for the purposes of statistical analysis.
Module 2 combines data exploration steps that include EDA, with uni-,
bi-, and multivariate statistical analysis, and examination of the most impor-
tant variables for the predictive model. Module 2 also combines missing
data and outlier analysis and temporal/spatial smoothing, which is often
overlooked in the data preparation phase. Techniques like data imputation
[e.g., multivariate imputation by chain equations (MICE) (van Buuren and
Groothuis-Oudshoorn, 2011)] or interpolation [e.g., regression methods
(such as LOESS/LOWESS), kriging, etc.] can be considered.