The flow pattern dependant mechanistic formulations found in the scientific literature for modelling the pressure gradient and liquid content of gas and liquid flow in wells have evolved considerably in the past twenty five years.
This paper tries to answer the question of how well the predictions of these models are able to match field measurements and how their performance can be measured and compared with other models like the proprietary model OLGAS.
The well cases used to test the models include diverse geometries, i.e. deep and deviated trajectories, and a range of fluid systems, i.e. gas-water, gas-condensate, oil-gas, and gaslifted oil production. These cases were obtained from data available from Alberta, Germany, and a sample of the Stanford Well Data Bank.
Software implementations of three mechanistic models were developed to be used as additional options in a commercial well simulator so that results could be compared on the same basis. The calculated versus measured pressure drops were transformed into nine statistical variables that were used to build an improved relative performance index to compute a new normalized index or grade that enhances previous published relative performance indicators.
In the course of this study, a modified version of the Gomez et al. model including a revised liquid entrainment correlation was implemented to overcome an apparent limitation in gas-lifted wells. The grade obtained with the model went from 30% to 70% for the subset of 34 gas-lift well cases while it went from 90% to 94% for all of the data set, making it the best published mechanistic gas-liquid flow model for vertical upward flow from among the three models benchmarked in this study.