Page 401 - Intelligent Digital Oil And Gas Fields

338 Intelligent Digital Oil and Gas Fields

to extract the maximum benefit from the enhanced monitoring capacity and
controllability of smart fields. The goal of such a system is in essence to con-
tinually maximize the life cycle value of the oil field by enabling proactive
optimization and decision-making, which in turn is enabled by real-time
monitoring, continuous model updating, and optimal control of the oil field.
Sarma also notes that the key components required to enable a closed-loop
optimal control system are as follows: a continuous data acquisition and inte-
gration system, a set of forward models relating the control variables to per-
formance indicators or objectives, ability to update these models with the
latest data, ability to optimize these models across multiple objectives and
constraints, and finally a system that enables easy consumption and imple-
mentation of decisions recommended by the system.
New sources of real-time data such as fiber optics and permanent down-
hole sensors have increased the volume of data collected by orders of mag-
nitude. However, much of these data are not used to the fullest extent
possible, and almost certainly not for proactive decision-making. For exam-
ple, pump-off-control and other well-related data can be used for predicting
the probability of well failure and thereby allow predictive maintenance.
However, maintenance decisions are still typically reactive: wells are fixed
after they fail. Such reactive solutions are in general “too little, too late”
and can be quite costly.
The second component of the closed loop, namely, the forward model,
has almost always been a reservoir simulation model in existing (partial)
implementations of the closed loop. However, the computational complex-
ity of simulation models, coupled with the significant time and effort to build
these models, makes the closed loop all but impractical to apply in anything
close to real time. Although techniques can be applied to alleviate some of
these problems, there is no complete solution today and this is undeniably
the most critical component and also the one hindering mass adoption of
closed-loop control.
The third component of the closed loop is technology to efficiently and
accurately update the forward model with the latest data. Techniques here
range from well-established deterministic gradient-based methods to more
recent stochastic/heuristic methods such as genetic algorithms. An addi-
tional aspect of the model updating problem is re-parameterization or
dimensionality reduction of the uncertain model parameters. While there
is still work to be done, Sarma feels that this is an area where significant pro-
gress has been made toward a practical solution that is applicable in a
closed loop.