PROPERTIES OF BLACK OILS - Petroleum Reservoir Fluid Property Correlations - page 42

Correlation equations with low values of ARE indicate that the
calculations describe a line that bisects the data. The relatively high
values of AARE for all correlations in table 3-7 are a reflection of the
precision of the laboratory-reported values of bubblepoint pressures.
Involved in this precision is the method of controlling and measuring
temperature during the laboratory procedures. A good correlation can
give results centered on the data, but a correlation cannot be more
precise than the underlying data.
Valko and McCain used essentially all of the data described in table
3-6. 8 Velarde et al. and McCain et al. each used approximately 40% of
these data, and Vazquez and Beggs used approximately one-third of
these data. 9 Hanafy et al. used less than 10% of these data to prepare
their correlation equations. 10
The Valko-McCain correlation equations appeared to give the best
results in the comparison of table 3-7; however, an examination of how
well these equations stand up across the data set is important. The 1,794
lines of data were sorted and sliced into 16 approximately equal subsets,
and the ARE and AARE for each of these subsets were calculated for
several correlations. Calculated values of ARE and AARE may be found
in figures 3-1, 3-2, and 3-3 for sorting on stock-tank oil gravity, solution
gas-oil ratio at bubblepoint pressure, and reservoir temperature.
The resulting graphs would be impossibly complicated if all
correlations were included. The Standing, Glaso, and Lasater equations
were selected because they seem to be the most popular in the industry
today. 11 The Labedi equations were selected because of the very low
ARE achieved on the entire data set. 12 Note that the results with the
Valko-McCain equations are consistently close to zero throughout the
full range of the data. The calculations with the Standing equations
are remarkably well behaved. The low ARE for the full data set for the
Labedi correlation equation is a result of the averaging of some large
negative values of ARE at one end of the spectrum of data with large
positive ARE values at the other end.
These graphical comparisons are not meant to indicate that one set
of correlation equations should be used for one part of the spectrum of
the independent variables and another set used for another part of the
spectrum. Rather, a single set of correlation equations should be selected
for use across the total distribution of possible independent variables.