This paper evaluates published correlations and neural-network models for bubblepoint pressure (pb) and oil formation volume factor (Bo) for their accuracy and flexibility in representing hydrocarbon mixtures from different locations worldwide. The study presents a new, improved correlation for pb based on global data. It also presents new neural-network models and compares their performances to numerical correlations.
The evaluation examines the performance of correlations with their original published coefficients and with new coefficients calculated based on global data, data from specific geographical locations, and data for a limited oil-gravity range. The evaluation of each coefficient class includes geographical and oil-gravity grouping analysis. The results show that the classification of correlation models as most accurate for a specific geographical area is not valid for use with these two fluid properties. Statistical and trend performance analysis shows that some published correlations violate the physical behavior of hydrocarbon fluid properties. Published neural-network models need more details to be reproduced. New developed models perform better but suffer from stability and trend problems.