American Oil and Gas Reporter - September 2017 - 52

proppant per foot for a 3,000-foot lateral.
Look-back analysis is a valuable management tool in shale analytics that seeks
to use field measurements to learn from
previous completion practices in order
to measure their effectiveness and refine
future practices. It is important to note
that since shales are well known for their
heterogeneous qualities, similar completion
practices will result in different well productivities based on the variances in shale
quality from location to location. Therefore, in order for this technique to work
properly, the reservoir quality of a given
well must be isolated (kept constant) during the analysis so that the analysis can

compare "apples to apples."
In look-back analysis, the predictive
model is integrated with Monte Carlo
simulation to evaluate the quality of the
completions and fracture stimulations.
Therefore, for each well, parameters that
represent reservoir quality (i.e., TVD,
net thickness, porosity, TOC, etc.) are
kept constant at the value measured at
the well, while the design parameters
(such as lateral length, clusters per stage,
fluid and proppant volumes, etc.) are represented by a triangular distribution using
the range from the dataset and the value
of the well as the most likely value.
Then, the predictive model is executed
1,000 times, each time a random combi-

FIGURE 2A
TOC Type Curves with 180-Day Cumulative Production
As Function of Lateral Length
1.5

1.5

140e3

180D-BOE(stb)

120e3

100e3

80e3
60e3

40e3

20e3
1,500

2,000

2,500

3,000
3,500
Lateral Length (ft)

4,000

4,500

5,000

FIGURE 2B
Job Size Type Curves
(Pounds of Proppant per Foot of Lateral Length)
1,000

400

1,500

2,000

120e3
110e3

180D-BOE(stb)

100e3
90e3
80e3
70e3
60e3
50e3
40e3
30e3
20e3
1,500

2,000

2,500

3,000
3,500
Lateral Length (ft)

52 THE AMERICAN OIL & GAS REPORTER

4,000

4,500

5,000

nation of the five design parameters is
selected to be coupled with the actual
reservoir parameters and presented to the
predictive model. At the end of each execution of the model, the result is the
production index (180 days of cumulative
production). Then the 1,000 production
indices for a given well calculated in this
way are plotted as a histogram. The resulting histogram demonstrates the potential production that may have been
achieved from each particular well given
its reservoir quality. Once the histogram
is generated, P10, P50 and P90 reserves
can be estimated for each well.
When the Monte Carlo simulation is
complete, each well's actual production
value is superimposed on the histogram
to identify its actual "Px" (cumulative
probability of the histogram, which determines the quality of the completion).
For this study, the Px values for different
quality of completions were assigned as
excellent (P20 and below), better than
expected (P20-P40), as expected (P40P60), worse than expected (P60-P80),
and poor (P80 and above).
For example, Figure 3 shows the results
of the look-back analysis for well 88-JC6. The P10, P50 and P90 for the 180
days of cumulative production values for
this well are 150,000, 130,000 and 95,000
boe, respectively. The well's actual production was 145,000 boe. Therefore, the
completion quality ranking of this well
was P15.
This look-back analysis was performed
for all 136 Marcellus wells. The final results show that 52 percent of the wells
were completed with "as-expected" quality, while the quality of 23 percent of the
wells exceeded expectations and 25 percent
fared worse than expected. After analyzing
more than 3,000 wells in multiple resource
plays, it appears that this distribution of
completion quality is common across all
of the plays evaluated, with at least 20
percent of wells ranking in the "worsethan-expected" category.

Completion Optimization
Another use of the predictive model
is incorporating it into an evolutionary
optimization routine for completion optimization. In this approach, the model's
input parameters associated with reservoir
characteristics are kept constant while
the optimization routine looks for and
evolves the most appropriate completion
strategy for a given well.
Figure 4 shows an example for a Marcellus well in Northeast Pennsylvania.
The predictive model includes reservoir
characteristics and completion/hydraulic
fracturing parameters from more than
400 wells. In this case, the operator was

American Oil and Gas Reporter - September 2017

Table of Contents for the Digital Edition of American Oil and Gas Reporter - September 2017