When no lab data is available, fluid property correlations are used to generate PVT inputs for engineering studies.
The issue is, PVT correlations are region-specific, and generic fluid properties simply don’t fit the bill.
Here’s a fantastic study by veteran RE Larry Wilcox on Malaysian oils using non-linear regression, non-parametric regression, and NNM (neural network modelling).
The study extended PVT data prediction to estimate oil composition, generate synthetic PVT experiments, create a matched synthetic EOS model, and use this model to replicate a real laboratory PVT study.
The match was within 3%!
Click on the image to view my LinkedIn post with a slideshow.
You can also read this technical paper that Larry wrote describing the study, which includes a description of non-parametric-regression and NNM.