Page 224 - Intelligent Digital Oil And Gas Fields

176 Intelligent Digital Oil and Gas Fields

To validate the output of the ANN, the workflow is preset with an inter-
2
nal correlation coefficient (R ) that rejects and accepts the computed liquid
rate or water cut. The automated workflow was proved in the following
events or production scenarios:
• Replicate and populate daily production data when pressure, tempera-
ture, and production signals are missing.
• Provide ESP production data when data transmission is frozen or elec-
trical power is shut down.
• Generated on-demand sensitivity analyses by changing pump frequency
and GL volume.
The prediction results were acceptable out for 20days (Fig. 5.13). The ANN
responds well within acceptable accuracy to changes in THP with pump fre-
quency and THP with gas-lift injection. The ANN was found to be an
excellent tool to populate missing data from production history and a useful
tool to provide on-demand sensitivities for changes in THP. The ANN pro-
actively predicts the liquid rate and water cut for the next 20days. However,
the ANN cannot predict water cut with acceptable accuracy; the water cut
results sometimes appear illogical or do not follow the water-cut history
trend. The main reason for this failure is that the incremental water cut is
not due to changes in frequency or gas volume but is more related to water
injection. We conclude that the ANN is powerful tool to be used as a VFM
to measure oil and gas, but should not be relied on to predict oil, gas, and
water beyond 20days. The ANN could predict short-term production
(<30days) with 90% confidence.

100
90
Correlation coefficient % 70
80
60
50
40
30
Oil rate correlation
20
10 Water cut corr
0
0 7 14 21 28
Days for prediction ahead of day 0
Fig. 5.13 Correlation coefficient for VFM based on an ANN. Oil rate can be predicted
with 90% confidence up to 20days. WC is unpredictable using an ANN.