A PVT Journey – Key learnings so far

As a practising Reservoir Engineer, I’ve always felt that I understood PVT. Like many, I use it almost daily, but somewhere deep inside I always realized that when push-came-to-shove, I’d struggle if/when asked to actually build my own equation of state models.

That realization, coupled with the knowledge that I’d soon be involved in several projects that would require just this, prompted the start of a 2 year+ journey of discovery, not just in PVT and EOS modelling but also with Python. You can read about the ‘end of the beginning’ of that journey in an earlier article here.

Before I go any further, I must first acknowledge some key people that have made this journey more enjoyable and productive.

Curtis Whitson and his team (Mathias, Bilal et al.) have been super helpful. Curtis co-authored the ‘Phase Behavior’ book that I used to cut-my-teeth with, is the industry go-to-guy on all things PVT, and together with his team have patiently answered many questions of mine over the last few years. Curtis also introduced me to the PhazeComp PVT Software, which leads me to the second key person. 

Aaron Zick – quite possibly the most knowledgeable person on the art of PVT modelling that I’m aware of. Aaron wrote / writes the fantastic PVT modelling package ‘PhazeComp’ and has always provided exceptional support and guidance. For those saying to themselves “What is PhazeComp ?” and fancy giving it a try, I would caution that the software is not for the faint-of-heart. No GUI’s or glossy manuals here, but if you are willing to put in the time to learn the craft and the tool, and absolutely, positively, need the Ferrari of EOS modelling (at a Toyota price), then accept no substitute!

In the last year and half, I have created and documented Single Carbon Number (SCN) and pseudoized EOS models for 6 fields, including saturated oil, lean gas and rich gas condensate systems. For your reading pleasure, here is a ‘top of the pops’ selection of key learnings so far.

Adjust critical parameters in a physically consistent way.

An EOS model behaves very non-linearly. If you materially change any one critical parameter in isolation, or begin with a poorly selected set of initial parameters then you can rapidly create an unphysical mess of a model. To keep critical parameters reasonably constrained, correlations such as the internally consistent Twu equations should be used for your starting point. Keep an eye out for monotonicity. Be wary about changing BIP pairs in significantly different ways or you can easily come up with a system with crossing Ki (equilibrium ratio) lines.

Specific gravity and boiling point temperature aren’t EOS critical parameters, so why are they measured and used?

Correlations such as Twu rely upon an understanding of Sg and Tb behaviour of your fluid to be useful, so make sure you use your distillation experiment results! Spending additional time up front to characterize your fluids’ Sg and Tb behaviour, along with its MW progression with distillation data and methods including Gamma regression will yield dividends. The more that your regression starting point reflects properties specific to your fluid, the easier your task will be.

Only regress to measured data, but include ALL of it

• Lab reported molar fractions aren’t normally measured, mass fractions are. Don't assume that your MW scheme matches the lab assumptions (normally Katz and Firoozabadi).
• Reported MW’s and SG’s often aren’t actually measured. Lab assumed MW and SG schemes for C7+ SCNs are often used to calculate these properties.
• Oil formation volumes aren’t actually measured, the oil volume at a given pressure is what is measured, and the last pressure point (atmospheric) is often arrived at via bleeding off pressure (a non-equilibrium method) meaning the residual volume – used to normalize all the reported measurements against – will always be slightly different to what is calculated with an equilibrated solution.
• GOR’s reported at standard conditions aren’t actually measured, they’re calculated at experiment conditions from to how much mass has been removed, relative to 'stock tank barrels' volume generally derived from a non-equilibrium process. Match to volumes and mass at experiment conditions instead.
• Gas viscosities aren’t normally measured, correlations are used.

Spend some time to understand what is being measured in your experiments and how, and regress against all of it – including data such as measured RFT pressure points or producing CGR trends vs depth if available. Don’t bother regressing against correlated or calculated results, embedded in which are lab assumptions about the fluid.

Interpolation over extrapolation

The goal is to create a model that can reasonably predict how your fluids will behave under a wide range of pressures and ratios. The wider the range of pressure and gas / liquid ratio data you can incorporate into the regression process, the more likely that your results will be approximately correct rather than precisely wrong.

Accuracy over precision

Don’t exclude that annoying outlier that is messing up your match quality unless you are certain there is a problem with it. Understand what predicted properties will most impact the answer to the questions that you hope to ask with this EOS model, and ensure those are most accurate, possibly at the expense of other predicted property accuracy.

Viscosity is not an EOS property

Viscosity is normally calculated with a correlation. In the case of the commonly used LBC correlation, it uses EOS calculated molar volumes to perform those calculations, but the correlation itself is not perfect, so don’t match viscosity at the same time as you regress the other EOS critical parameters. Finalize your EOS critical parameters regression first, then - with those properties fixed - you can perform a second regression on measured viscosity by changing Vcvis or Zcvis as well as the higher order LBC coefficients. To improve your viscosity predictions at surface conditions, incorporate Orrick & Erbar correlation predicted viscosities for your SCN species during this regression.

Don’t just guess pseudo lumping schemes

The selection of the ‘best’ pseudo scheme for lumping your high-resolution EOS model has long been subjective, with rules of thumb and some prescriptive guides often being applied. By using an EOS tool that can be programmatically launched, you can actually explore the full subset of ‘reasonable’ combinations. I use a custom Python script linked to PhazeComp to explore many tens of thousands of combinations for a given target model size so that (a) I can be sure that the final scheme is optimal and (b) when repeated over different model sizes you can quantify the smallest lumped scheme, beyond which unacceptable quality degradation begins.

Select your initialization molar splits carefully.

Often the subjectively selected ‘best’ sample is chosen to represent insitu molar composition without much additional thought. It can pay to investigate this a bit more rigorously. An example of this might be taking each initial sample candidate as the starting point for a succession of gravitational segregation experiments to describe the initialized compositions that would result throughout the structure, then taking virtual samples of those compositions at all other DST well-test or initial production depths (if available) to see which one delivers the most internally consistent observed GOR behaviour. If you have a GOC in your field, ensure that it flashes at the correct depth, which might entail slight methane or heavies composition tweaks.

Wrap up

If you have read this far, then many thanks. I’d encourage any young engineer out there (or old ones for that matter!) with an inkling of curiosity on this topic - but not sure of where to start - to simply jump in. Grab books on the topic, get some lab reports and modelling software and just start working through it - it won't happen via osmosis.... Develop contacts with people more knowledgeable than you on the topic and willing to share that knowledge and cherish them.

I’ve thoroughly enjoyed this trip so far and am looking forward to continued learnings in the future.

Mark Burgoyne