Automatic History Match and Upscaling Study of Vapex Process and Its Uncertainty Analysis
Abstract
Vapour Extraction (VAPEX) is a process to recover heavy oil by injecting vapourized
solvent into a reservoir. In order to ensure successful commercial application of a
VAPEX process, reliable prediction of VAPEX performance is crucial. The current
practice for VAPEX performance prediction is using analytical scale-up methods that
translate laboratory results to field applications with analytical models. However, the
drawbacks associated with the analytical scale-up methods are that they only consider
single phase flow and gravity drainage and cannot take reservoir heterogeneity into
account, which limits the applicability to real field cases. In this work, an effort was made
to investigate the capability of predicting up-scaled VAPEX performance through
numerical simulation.
In this study, numerical simulation was conducted to up-scale 2-D VAPEX tests and
to predict 3-D VAPEX performance. The 2-D VAPEX test was conducted under
conditions very close to those for the 3-D test which is done by the Saskatchewan
Research Council (SRC). In each test, the initial waterflooding was conducted prior to the
subsequent solvent injection. Then, a numerical model was established to simulate the 2-
D test. History match of the 2-D test was conducted by tuning the uncertainties, such as
the relative permeability, capillary pressure, solubility, and the wall effect in sandpacking.
Afterwards, the tuned parameters were applied to predict the 3-D test
performance. Through comparison of the predicted and experimental results in the 3-D
test, the capability of predicting an up-scaled VAPEX process through numerical
simulation was examined, and the differences between the physical and numerical
modeling were identified.
As history matching of VAPEX experiments is a complex, highly nonlinear, and nonunique
inverse problem, a modified genetic algorithm (GA) was developed to assist with
the history matching process. A population manipulation database and artificial neural
network (ANN) were incorporated with GA to enhance the computational efficiency and
enlarge the search range. Compared to conventional GA, the computational time in this
modified GA approach was reduced by 71%, and an excellent match between the
simulation data and experimental data was achieved.
The upscaling study results show that the numerical method used, compared to
analytical models, has greater potential to be used as a scale-up method because of the
improved prediction results. Due to the nature of the numerical simulation method, it is
difficult to match the early stage of the solvent injection process, which results in the great
uncertainties. It was demonstrated that the waterflooding performance can be successfully
predicted, whereas the uncertainty in scaling up the VAPEX process is great. In the
waterflooding period, the predicted oil recovery factor was 25.8% compared with 23.4%
in the 3-D test. In the VAPEX process, the difference between the predicted and
measured oil recovery factors was in the range of 0.8–25.1%, depending on the different
combination of uncertain parameters. This fact indicates that more work on this topic is
required to reduce the uncertainties in predicting the field-scale VAPEX performance,
and it would be especially valuable for field applications.