This paper presents a methodology to answer quantitatively the question of which type of steady state point flow model implementation suits a particular production system better. The statistical tools used are publicly available and the only assumption is that sufficient, reliable, field data, and a simulation program with multiple point model implementations are available. A concrete example is worked out to illustrate the analysis.
Thirty five proprietary and published gas-liquid flow point models were used in a pipe flow simulator with the aim of statistically confirming clusters of common modelling technology. The experimental pressure drops and liquid holdup were published by Lagiere et al. in 1984 on a total of 19 runs on a 361-km long subsea pipeline joining the, now decommissioned, North Sea offshore production platform of Frigg to the terminal at St Fergus, in the UK
A data set with 2541 observed and simulated data and 23 variables was created using the R scripting language with factors such as test section, elevation profile (coarse or detailed), and boundary pressure location (up or downstream). The result output text files and the R code are publicly available online. Hierarchical clustering and K-means cluster analysis were applied to test what models consistently rank high for pressure drop prediction. The Relative Grade was used to identify the most robust models that appear consistently within the top ten for all the simulator configurations used. This measure of performance uses four different formulas of experimental error, four of variance, and one of model inadequacy.