The predictive accuracy of eight critical two-phase flow correlations aretested against field measured production data, from 210 well tests, coveringbroad range of production rates, choke sizes, upstream pressures, gas-liquidratios and oil API gravities. Test data are divided into four selectedcategories based on choke size (D): D < 6, 6 ≤ D < 10,10 ≤ D < 30 and D ≥ 30/64.
The correlations included in this study are those of Gilbert, Ros, Baxendall, Achonge, Poettmann-Beck, Omana, Ashford and Hazim-Ghassan.
The average percent error, absolute average percent error, and standarddeviation are computed for each correlation. Due to inconsistency of resultsobtained by the included correlations, the multiple regression analysis is usedto find out correlations that best fit the measured data, as a result, four newcorrelations are developed (a correlation for each data category).
Based on the statistical results, the new correlations clearly outperformed theremaining correlations.
Several correlations have been published for describing critical two - phaseflow through wellhead chokes. Most of these correlations were based on limitedranges of flow variables. Their validity is limited by the quality and scope ofthe data upon which they are based.
The objectives of this study are:
To indicate the accuracy of the existing critical flow correlations.
To modify the existing correlations if necessary or develop newcorrelations that best fit the measured data.
The correlations compared are those of Gilbert1, Ros2, Baxendall3, Achonge4, Poettmann Beck5, Omana6, Ashford7 and Hazim-Ghassan8. Some ofthe correlations require values of fluid physical properties, the following PVTcorrelations are used to estimate them:
Fifty six tests from Iraq wells, 108 from Poettmann's study, 27 from Ashford'sstudy and 37 from Omana's study are combined in a data bank for thisinvestigation.
Each of the well tests is examined for unreasonable values by comparing themeasured production rates for a well test with the production rates predictedby all correlations. As a result, eighteen well tests are removed The existingrevised bank now contains 210 well tests. The ranges of flow variables coveredby the well data are given in Table 1.
The final equations of the compared correlations are given below (a detailedanalysis If the correlations may be found in the original sources in thereferences)
The equations proposed by Gilbert, Ros, Baxendall and Achong, are all of theform:
Equations (available in full paper)
It is very important to use the absolute average percent error (AAPE) as a baseof comparison instead of average percent error (APE), because when the laterused, the negative errors cancel the positive errors and thus the APE willappear small. Using the AAPE removes this cancelling effect by transforming allnegative error to equal magnitude positive errors. This is illustrated in Fig.1.
