![Life of Field Cooldown Time
Oil Rate
Phi
Beta
Kappa
Year
Upside
Most Likely Temps
Cooldown Time
Downside
Target Cooldown Time [18hrs]
What’s the chance that all wells are on their downside? Year
6
What’s the probability that the cooldown time will be less than 18hours?](https://image.slidesharecdn.com/sessionivwatsonintegratingreservoiruncertaintiesintoflowassurancestrategies-13318876203736-phpapp02-120316035113-phpapp02/85/Integration-of-Reservoir-Uncertainties-into-Flow-Assurance-Strategies-6-320.jpg)
Integration of Reservoir Uncertainties into Flow Assurance Strategies
- 1. Integration of Reservoir Uncertainties into Flow Assurance Strategies Dr Martin J Watson SPE Applied Technology Workshop Bridging the Gap Between Reservoir Engineering and Facilities Design 14th&15th February 2012, San Antonio
- 2. Facilities Thermal Design & the Reservoir Many uncertainties in the thermal hydraulic design of Facilities lie in the reservoir How much is going to flow At what temperature The traditional approach to Facilities design is to design to a FWHT and a “design rate” But such decisions are arbitrary Over conservative for some, under conservative for others E.g. What FWHT is conservative for both wax and corrosion management? Thermal Hydraulic IPM can help manage these risks more rigorously Rigorous thermal hydraulic model from reservoir to processing facilities Can investigate how the reservoir uncertainties effect Facilities Design Without the need for arbitrary intermediate boundary conditions What follows is an example in hydrate management But equally applicable to wax, corrosion, chemical injection, etc 2
- 3. Hydrate Management for a Deepwater Oil System Phi, Beta, Kappa fields 7 production wells planned 25mile daisy chained tieback in deep water 3
- 4. Deepwater Oil Hydrate Management & Cooldown Time 350 Hydrate Require a good cooldown Dissociation time as a hold time before 300 Curve the expensive hydrate Hydrates No Hydrates management operation 250 needs to be carried out Typically >10hours is Pressure (bara) Normal required, making most Hot Operation 200 shutdowns recoverable Restart with a Hot Restart Shut in 150 Cooldown 100 Blowdown & Dead Oil Displacement 50 0 0 5 10 15 20 25 30 35 40 45 50 Temperature (°C) 4
- 5. Life of Field Cooldown Time The Cooldown Time of a production system changes through life As the watercut, GOR and relative rates from each well changes On the FEESA website is described a method for turning a life of field thermo hydraulic simulation and a transient simulation into an estimate for cooldown time throughout life http://www.feesa.net/consultancy/casestudies.html Based on Newton’s Law of Cooling Min Pipeline Temp at Steady State Tmin SS − Tamb t = Bln Ambient Temperature Thyd − Tamb Cooldown time Hydrate Avoidance Temperature Fitting parameter 5
- 6. Life of Field Cooldown Time Oil Rate Phi Beta Kappa Year Upside Most Likely Temps Cooldown Time Downside Target Cooldown Time [18hrs] What’s the chance that all wells are on their downside? Year 6 What’s the probability that the cooldown time will be less than 18hours?
- 7. What is the probability the cooldown time is <18hours? Make some assumptions about the probabilities of the uncertainties I.e. Probability of upside, expected and downside, PI and Reservoir temp Probability density function Generate multiple combinations scenarios who’s probability we can calculate E.g. Kappa and Beta on expected, Phi on downside, etc We could simulate each combination to calculate cooldown time Generate a distribution of probability vs cooldown time However, there are too many possible scenarios Even if there we assumed there was only upside, expected and downside; for six uncertainties (PI & Tres for three reservoirs) 36=729 combinations 729 life of field simulations is impractical, even with the fastest thermal hydraulic simulator In reality there are many more We needed to choose the least number of runs to yield the most information about how these uncertainties affect cooldown time 7
- 8. Design of Experiment (DOE) Design of Experiment (a.k.a. Experimental Design) is the process of reducing the number of experiments whilst still ensuring statistically meaningful results Various Experimental Design methods in the literature Most famously the “Monte Carlo” method This requires too many simulations to be run (tens of thousands) We used the Taguchi method Developed to improve product design and manufacturing methods Procedural In this case we used Taguchi to reduce the number of simulations to just enough to develop a correlation for cooldown time versus our uncertainties Which we can then use in a Monte Carlo Simulation A reference Manivannan, S. et al, 2010, Taguchi Based Linear Regression Modelling of Flat Plate Heat Sink, J Eng & App Sci, Vol 5, Issue 1, 36-44 8
- 9. Applying the Taguchi Method to this Problem Requires the user to do sensitivity studies Rate the uncertainties in order of most important to least Phi PI, Phi Tres, Kappa Tres, Beta Tres, Kappa PI, Beta PI The rest is procedural In this case, there are 6 variables, a L8 orthogonal array was used Selects 27 runs from the 729 possible Phi Kappa Beta Phi PI Temperature Temperature Temperature Kappa PI Beta PI Run 1 0 0 0 0 0 0 Run 2 0 0 -1 -1 0 1 Run 3 0 0 1 1 0 -1 Run 4 0 -1 0 -1 1 -1 Keys Run 5 0 -1 -1 1 1 0 -1 Downside Run 6 0 -1 1 0 1 1 0 Most Likely Run 7 0 1 0 1 -1 1 1 Upside Run 8 0 1 -1 0 -1 -1 Run 9 0 1 1 -1 -1 0 Run 10 -1 0 0 -1 -1 -1 Run 11 -1 0 -1 1 -1 0 Run 12 -1 0 1 0 -1 1 Run 13 -1 -1 0 1 0 1 Run 14 -1 -1 -1 0 0 -1 Run 15 -1 -1 1 -1 0 0 Run 16 -1 1 0 0 1 0 Run 17 -1 1 -1 -1 1 1 Run 18 -1 1 1 1 1 -1 Run 19 1 0 0 1 1 1 Run 20 1 0 -1 0 1 -1 Run 21 1 0 1 -1 1 0 Run 22 1 -1 0 0 -1 0 Run 23 1 -1 -1 -1 -1 1 Run 24 1 -1 1 1 -1 -1 Run 25 1 1 0 -1 0 -1 9 Run 26 1 1 -1 1 0 0 Run 27 1 1 1 0 0 1
- 10. Method Each of the 27 Maximus runs gives us a CDTmin. We then use this data set to fit a correlation 2 2 CDTmin C0 aXn bXn 1 ... eXn fXn 1 ... R C0 = Constant The values of these are fitted by regression, e.g., Italics a, b, c etc. = Coefficients in Excel Xn = The outcome of variable n, taking the value of +1 or 0 or -1. Variable n refers to the set {Phi temperature, PI…etc.} Xn2 = Square of Xn hence can only be 1 or 0 R = Residual term, assumed negligible The effect of these parameters on CDT can now be predicted without running a full simulation by entering +1, 0, -1 into the equation 10
- 11. Comparison of Correlation vs Simulation 40 35 30 25 CDT (hrs) 20 15 10 5 CDT calculated from Maximus results CDT calculated using polynomial 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 Run number The average difference was 2% This was deemed acceptable, given the assumptions about the uncertainties This correlation was then entered into Crystal Ball for the Mote Carlo Simulations 11
- 12. Results Conclusion; given the reservoir uncertainties, the P90 CDT is 18hrs 90% chance that the CDT will be equal to or greater than 18 hrs throughout field life On Normal Operation 12
- 13. Other Uses - MEG Injection System Sizing Montini et al (2011) “A Probabilistic Approach to Prevent the Formation of Hydrates in Gas Production Systems” ICGH 2011 (Edinburgh) West Nile Delta Project Large offshore gas condensate network 34 wells, ~380km of pipelines Traditional design approaches would lead to it being the world’s largest MEG injection system! All the worst case scenarios occurring at once The Operator sought an alternative approach! 13
- 14. A Probabilistic Approach to MEG An alternative approach to the traditional worst of the worst of the worst…… A life of field thermo-hydraulic compositional evaluation Proved that many of these worst cases can’t happen in the same year A statistical investigation Reduced the MEG design rates by a factor of four Still 99% confident that hydrates are avoided Why not 100%? It takes hours to form a hydrate blockage in such systems Remediation is possible Scope for future work A better quantification of the risk of forming blockages A better cost estimate of remediation A better justification for “99% certainty” But BP were happy Help make the project feasible 14 Moved into FEED
- 15. Conclusion Facilities Engineers have a lot to learn from Subsurface Engineers! Life of Field Approach Viewing issues on a life of field basis rather than “mass sensitivity studies” Easier to understand the problem, easier to explain to others Integrated Production Modelling Removing arbitrary and erroneous boundary conditions (e.g. FWHT) Investigating the impact of Reservoir Uncertainties A statistical approach to uncertainties Rather than assuming all worst cases happen at once Build a picture of what is statistically likely to happen “We believe this is good for 90% of all scenarios we expect, and have a plan for the other 10%, do you want to go ahead or not?” On real Flow Assurance projects, such methods have; Reduced the number of excessive design margins Have proved conventional (proven) insulation systems could work, where traditional design methods pointed to less proven low U value solutions Made MEG injection practical for a large gas network 15