Page 118 - Intelligent Digital Oil And Gas Fields

86 Intelligent Digital Oil and Gas Fields

• Level 1: Simple averaging (summarizing).
• Level 2: Extrapolation, data follows trend and tendencies.
• Level 3: Replace by data-driven analytics, for example, artificial intelli-
gent components (Chapter 4).
• Level 4: Replace data through physics-based calculation. Using an engi-
neering model-based physics to replace pressure in build-up time or
to replace flow rates data using virtual metering system (covered in
Chapters 5 and 6).
Our experience in several projects is that replacing data with models often
presents challenges (Al-Jasmi et al., 2013a,b; Rebeschini et al., 2013). For
example, in one water flood that had multiphase flowmeters (MPFMs)
and real-time artificial lift data, we used different processes to aggregate data.
These included artificial neural networks (ANN), fuzzy logic (FzL), well-
performance evaluation (analytic flow analysis), a one-dimensional (1D)
analytical model, and 3D numerical models. We learned that there is not
a single, reliable, and confident data aggregation process for all cases.
A fit-for-purpose model should be determined for each situation and can
depend on the reservoir drive mechanism, state of flowing condition and
flow regime, artificial lift type, and fluid types.
In the water flood example, we found that using ANN to estimate short-
term production was effective but that predicting water cut or oil rate after
water breakthrough could introduce errors. The best approach is to train the
ANN with a physical model. A 3D physical model cannot be used to replace
real-time data directly (month vs. minutes), but engineers can understand
that training the ANN to predict the water breakthrough depends on his-
torical data and reservoir properties. The most important factors are:
• Wells producing below or above dew or bubble point pressures.
• Water from water flooding or an aquifer breaking through into the wells.
• Reservoir flow regimes (radial, bilinear, transitional, boundary
dominated).
• Flow conditions (steady, pseudo-steady, and -unsteady states).
• Well loading up or well flowing in critical conditions.
Fig. 3.3 shows eight different cases to replace or fill in real-time data. Pres-
sure, gas, or oil rates and water cut are shown versus a period of 7days. The
replaced data is shown by dotted lines. The plot describes the following:
• Fig 3.3A shows a well producing under pseudo-steady-state condition.
The pressure depletes (dp/dt¼constant), the water rate is near constant,
and the total gas rate declines with time. The rate data gap could be
reasonably replaced by using simple statistic (averaging).