Eclipse Convergence problems

[halim78]

Hello,
I'm running a forecast case using Eclipse and I have a lot of convergence problems.
Can any body help me on this issue or give me some books on fixing these problems.

Here is a snapshot of one of the problems:

[temr]

There is no book where you can find answers for non-convergence ( as far as i know) in eclipse
This problems accures when eclipse can not solve the non linear equation - for this ( how its culculates you can find a lot of books here).
But theres is some points you should aware of when building you sim model which effects the speed of calculations.
1 - cells size and shape - try to take out thin cells form you model
2 - cells poro and perm - try to avoid rapid changes of perm and perm from cell to cell - less stochastic more deterministic algos for property modeling or use smooth property in petrel
3 - try to not connect different PVT regions
4 - before running the models with wells run the model without wells let say for 5-10 years . After the run upload the sw property and use it as initial Sw for well run in this case you will get true equilibrium and you will avoid problems with cross-flows
5 - look for you BHP of you wells , some times run can be stop if eclipse can not eject the history volumes to the cells
6- faults shape - try to make zigzag fault in the sim grid

[itag]

Tem's advice is good.

In many cases the two most common problems are:

a) Low porosity and high permeability cells - these give rise to througput problems. Indentifying and removing these cells ( or lowering the perm there) can help. You need to be able to plot permx vs poro to find these cells.

b) Often too high a vlaue of PERMZ can be a problem. Plotting the histogram of TRANZ or PERMZ and using a MAXVALUE keyword to limit the highest 10% can often improve convergence while not affecting the solutution. ( similarly reducing kv/kh oftne improves convergence - but now the answer will most likley change.

There are many other reasons. I've found these 2 are a good place to start.

itag

[halim78]

Thank you very much for these information.
All these points are Ok in my model, I think the problem is relative perms curves which have some irregular shape, I'll fix them and see the results.

[hkazem]

Dear Temr,

What would be a solution if there is "cross flow" in the reservoir while initiallizing the simulation model? What parameters you usually check in this case?

[temr]

Hello hkazem.
The solution is quite easy.
Make run for let say 50 years without the wells
After run export swat from restart file and use it as initial one.
Time machine )))
Read carefully what ve been written above

[dipak_m]

Dear Temr,
Please elaborate in details when I run the model for 50 years without wells and use as a initial saturation to avoid cross flow.

[vinomarky]

The process that temr is illustrating is one where your model (as originally initialized) may not be in equilibrium - consequently from the very start you have fluid wanting to move around without even any well penetrations. By allowing theses fluid movements to fully equilibrate over time and using the resultant (settled) saturation grid as your new initial saturation grid it can remove convergence problems in some cases.

In my experience though this is only really needed when people try to initialize with scenarios like 'tilted contacts' without incorporating the physical cause of the observation.

Another thing to look at (along the lines of what the guys above point out) is to see if there is consistency in the reported blocks in which the convergence problems are occurring it - it may be as simple as zeroing out a handful of problem cells.

If instead it is indeed down to abrupt changes in your rel perm curves, then you've probably addressed the problem already.

[dipak_m]

thanks

[temr]

It is easy
in schedule section just put
Dates
like
.....

SCHEDULE

dates
1 'MAY' 1988 /
/

dates
1 'JUNE' 1988 /
/

.
.
.
.
dates
1 'JUNE' 2050 /
/

END

Again this method will avoid all possible block crossflows

[dquento]

hello all,
i have encountered a series of problems during a simulation run similar that mentioned in this thread and all related to WELL 3-113. i generated Gas PVT data because it wasn't available. only one PVT region in the model.
Problem 1/3/1990 4230.000 UN-PHYSICAL VALUE FOR 1/BG IN PVT TABLE 1 WELL 3-113 PG = 5.0712E+00 RV = 0.0000E+00 1/BG = -4.0867E-01
Problem 1/4/1990 4261.000 UN-PHYSICAL VALUE FOR 1/BG IN PVT TABLE 1 WELL 3-113 PG = 3.8311E+00 RV = 0.0000E+00 1/BG = -1.7119E+00
can someone please explain to me what this means?
thanks, regards.

[temr]

It is mean that during extrapolation of Bg values outside pressure nodes you gave in PVT section eclipse gets negative compressabilities which are non physical
this usualy happens if BHP of the well drops down dramatically ( 1 bar ) or increases (injectors)
look for the thread in the forum excell file to check negative compressabilities