Reservoir Simulation - Science topic

First create a numpy array of x y location you want to try out, secondly the well location description in the eclipse data file, cut that section and save in a file , call it change.dat. In the main eclipse data file change that portion and write “INCLUDE” then under it put the “change.dat”. This will make eclipse read that well property as an external file via the include statement. Now write a simple python code using freas and fwrite commands that will take the n by 2 numpy array you created earlier. In a for loop, take each row of the n by. 2 array, use open/fwrite commands on the change .dat file so you can be over writing the X and y location in that file. Run close by invoking “ecl XX.dat” where XX.dat is the main of your eclipse data file. Make sure you use the EXCEL and SERPARATE key words in the data file to output the oil/water/gas production rate to compute your NPV. You can use reinforcement learning to solve for the steepest ascent value of this NPV taking into account the discount factor and oil price per barrel and also cost for disposal and water treatment. You can let RL guide this well locations automatically instead of using brute force.at the end the algorithm will converge on the optimum x and y location with the maximised NPV value. Hope this helps.

Mohammed Qassim

Thank you for your question. Yes, it is possible to use Petrel in conjunction with Python for various purposes, including creating plugins and custom workflows. While Petrel does not have a direct Python integration like it does with MATLAB or C# in Visual Studio, there are alternative approaches you can consider.

1. Ocean for Petrel: Ocean for Petrel is a software development framework provided by Schlumberger, the company behind Petrel. It allows developers to extend the functionality of Petrel through custom plugins and workflows. While Ocean for Petrel primarily supports .NET languages (such as C#), you can still leverage Python in combination with .NET by using IronPython, which is an implementation of Python that runs on the .NET framework. IronPython allows you to write Python code that can interface with .NET libraries and, consequently, interact with Petrel through the Ocean for Petrel framework.

2. Python.NET: Python.NET is another option that enables Python and .NET interoperability. It provides a bridge between Python and .NET, allowing you to use .NET libraries within Python code. With Python.NET, you can potentially create plugins or custom workflows for Petrel using Python as the programming language. This approach offers flexibility by leveraging the power of Python while integrating with Petrel through the .NET framework.

Both the Ocean for Petrel framework with IronPython and Python.NET offer possibilities for using Python in conjunction with Petrel. They enable you to access Petrel's functionalities, develop custom workflows, and create plugins using Python code.

It is worth noting that the specific implementation details may depend on the version of Petrel you are using, as well as the requirements of your project. Therefore, it is recommended to refer to the official documentation and resources provided by Schlumberger to ensure compatibility and obtain detailed guidance on integrating Python with Petrel.

I hope this information helps you explore the options for using Python with Petrel for creating plugins and custom workflows. If you have any further questions or need additional assistance, please feel free to ask.

First several setups might exist. In some of them (e.g., so called PetrelStudio), the data is not in the Petrel project but rather in a database elsewhere and the project only reference data with or without cache. In the following I will focus on (quite common) case in which the data is in the Petrel project. In this case, the file to data correspondance is not transparent nor bijective. The formats are not explicit and not straightforward. You will have to access via Petrel and export your data. There could be exception in which you might be inposition to reverse engineer format and file allocation, but this would typically be very rare exception and unreliable. If the problem you face is lack of access to petrel, you are in a bad spot. If it simply not knowing how to do it, then it usually takes a bit of trial but export is typically doable, reasonnably trusworthyt for accuracy and completeness and can be automated. It certainly is possible to transfer most of the data to something that can bea read in CMG suite with anot too much (but not zero) reformatting efforts. Am I clear?

Aahed Alhamamy Thank you for the answer, it really deepens my understanding of the software. And indeed that there are only a few tutorials on YouTube.

I believe that I need some data set as you mentioned because from one tutorial I watched, he used the data set but didn't explain anything about it.

However, I am not familiar with the data set for SGSIM, can you provide me suggestions regarding this issue?

Dear Fatemeh Ghallath

There is a good article that will benefit to you

If you need the PDF just tell me, also I attached an article that seems good

Dr. Ahmed

Mehdi Razavifar

Hello Mehdi, have a good day

Maybe the following links can help you:

Thank you Mehdi Razavifar jan.

Hi Erfan Rahimi , in time-dependent 3D geomechanical simulation ("4D" coupled flow and geomechanical simulations) there a many interdependencies between material properties (e.g., elastic properties, strength properties, porosity/ permeability, fluid properties) and simulated fields (stress, strain, pore pressure).

You have to think carefully when including additional interdependencies (or vice versa NOT including these interdependencies), whether they add (i) a lot of additional complexity, (ii) create a lot of additional insight, (iii) create complexity without creating insight, (iv) the error incurred by failing to include the interdependency. As you correctly point out, updating elastic properties in the overburden is NOT commonly for coupled flow and geomechanical modelling. This is one of the cases where you make the simulations a lot more complex, without adding a lot of insight. Elastic properties due to stress/strain changes in the overburden change by less than a percent from their initial value - and updating the elastic properties will affect the simulated stress field by an amount which is insignificant compared to our ability to calibrate the stress field. In your picture, you show a loop which includes updating of velocities for 4D seismic attribute generation. Here updating the velocities (even by less than 1%) results in something we can observe in field data in the form of time-lapse timeshifts. In a similar manner, if permeabilities are stress dependent in a significant manner and neglecting to include this coupling will create a large error, it is customary to include this coupling. Another example of coupling which is sometimes, but not always, used is to use non-linear stress-strain relationships in the reservoir, if significant compaction occurs and the reservoir rock will experience irreversible compaction.

In summary, keep models as simple as possible, and add complexity if there is a good reason. Do not fall into the trap of making models "complex" for the sake of complexity. Complex models are harder to interpret, and don't necessarily provide more insight. There is an unfortunate tendency of assuming that "complex" models are "better" models. They sometimes are, and often are not.

Hope this makes sense, and addresses your question.

Cheers,

Jorg

The costs involved in building, calibrating and using other types of forecast models (3D reservoir dynamic simulation models, data driven machine learning models) are such that a large volume of reserves (most of the mature fields and a large fraction of remaining reserves) are derived from DCA. One can however surmise that streamlining of machine learning usage will enable such approaches to overcome simpler form of data driven approaches like DCA. In my opinion, these approaches will not overcome 3D reservoir models for green (or immature) fields any time soon.

Reservoir temperature in accumulations depends on their occurrence depth and geothermal specifics of the giver Earth crust area. Observed temperatures vary from near 0°С in gas-hydrate accumulations to hundreds °С in deep-lying formations

Hi Jungin! This is the saturation profile I have applied.

Pierre Bergey

Hello,

I want to add monitoring pressure gauges in my CO2 injection model (E300). With the injection I want to monitor the pore pressure increase at particular depth or grid?

Any recommendations. Thank you.

Domagoj Vulin Structured grids for ANSYS CFD solver simulations are commonly build with ICEM/CFD Hexa, if they are preferred or required.

Regards,

Dr. Th. Frank.

Hassan Mahani I saw you mentioned one the works of Alan S. Emanuel from Chevron. Back in 1970s, Emaneul had published remarkable works in the area of commercial simual. In fact, in the first year of my Ph.D. I was totally stuck, and I only figured out the general idea of my PhD work after reading one of his papers ( )

It was interesting to me to see his name here. If I am not wrong Emanual has around 30 papers in SPE library.

The fact that there is no gas in your system (meaning low compressibility) and the small description you provide suggest that your are dealing with a small closed (no analytical or explicitly gridded aquifer) low compressibility reservoir. In such situation and unless you have setup a pressure control of your injection or production or both, seeing strong pressure variations is normal. You would need to make the injection or production or both controlled by a regional pressure.

There is a major issue with FZI method. In a 2018 paper it was shown that FZI is strongly dependent to grain size rather than pore space. Fluid flow is through pore spaces and not through grains. Thus FZI, can't characterize flow units very well, especially it fails in carbonates and dirty sansdones.

I have seen many papers and people claiming that FZI can identify flow characteristics of rocks. However none of them proves this claim.

They simply plot RQI vs phi(1-phi) and obtain some HFUs. Ok. What is common for rocks in each HFU? Pc? TEM(true effective mobility)? Kr? they never show this.

I am sure that if in your study you check the relationship between HFU and other parameters such as Pc and TEM, you will find no relationship, specifically for carbonates and dirty sands.

What is the solution?

This solution is simple. Darcy's law (when apparent velocity is replaced by true velocity) shows that the parameter that defines a flow unit is K/phi (permeability over porosity). so simply plot k vs. phi and rocks on a straight line form a HFU. This led us to propose a correct method to find HFUs called FZI* or FZI-star.

see more at

and Rasa Soleimani I have tried your respective codes but they seem not to be working. This is my code but it is not working as well. can you assist you making it work or any suggestions on what I need to do? My Eclipse version is 2010.1.

% @echo off

system('call C:\ecl\home\$eclrc.bat');

system('set ECLVER=2010.1');

system('set TEMP_PATH=%PATH%');

system('set PATH=C:\eclipse-2010\2010.1\lib\pc;%PATH%');

disp(' ----------------------------------------------------------------');

disp(' 01/01/1990 2:08:14 PM');

disp(' ECLIPSE version 2010.1 dataset c:\Eclipse-2010\home\run_p');

disp(' ----------------------------------------------------------------');

system(['C:\ecl\2010.1\bin\pc\eclipse.exe', EclipseInputFile]);

system('set PATH=%TEMP_PATH%');

system('set TEMP_PATH= ');

end

I await your urgent help

one of the challenges in the reservoir simulation is the reservoir characterization especially in terms of vertical rock facies prediction which are the input for the spatial modeling of facies .the good prediction of facies will improve the porosity-permeability relationship and help in getting accurate permeability model that help to obtain a realistic reservoir model.

Munqith, if you have production breaks the water/gas segregation will cause fluctuations. You have similar early on 2018 also. LGR around the wells may mitigate some of these effects. If you have hysteresis option turned on, some of these fluctuations maybe due to the relperm change between imbibition and drainage cycle.

OpenGeosys

https://www.opengeosys.org/

Hi, to quantify the residual oil. We need to first know if hysteresis is applied in the simulation model. If no hysteresis is applied, then residual oil is the Sorw, which you can read from the O/W Kr curve; If hysteresis is applied, then residual oil is dependent on the initial water saturation, which can be obtained from hysteresis model, then each cell of a certain satnum and Swi can have its own Sor.

The problem with the weather is that we know the system well enough but we can't measure the state variables accurately. It is chaotic so we use an ensemble approach. In Petroleum, the behaviour of the system is stable but we don't know the system, again due to lack of information. We have some hard data but it's often not representative (e.g. core measurements apply to a tiny volume). We have soft data like seismic which is an average over certain ranges. Then we have the unknowns (some we know we don't know and some we just don't know if they exist - e.g. sub-seismic faults and other barriers). So there is a LOT of uncertainty and that can outweigh the numerical approximations by far. There is no such thing as the "right model" and we need to consider many models.

From a reservoir / recovery perspective one can consider that infinitely small sized slugs tangent the behavior of simulatenous injection. But this is not true from an operations and design of facilities perspective. Equipments and procedures for true simultanenous operations will differ from those of alternating with short slugs (CAPEX /OPEX will be different, HSE aspects too).

For the second question, there are many possible answers. I would structure things considering different configurations. First configuration would be reinjection of associated gas (minus whatever was used for compression and other utilities) eventually focusing WAG on part of reservoir(s), second configuration would be maximum efficiency level of gas injection (no limit on gas availability).

Study on Mechanisms of sc-CO2 Injection as Tertiary Oil Recovery Technology for Carbonate Reservoir/Rocks

Ikechi,

What are the parameters that you are trying to match? Do you have observed data for history matching, such as rates and pressure?

Thank you Haw and thank you Arshad for your answers.

Industrial grade reservoir simulators like IMEX/ECLIPSE/INTERSECT/... are probably unsuited for the type of work you want to perform. You likely need to access and tweak the physical engine of the simulator. R&D codes are what you are looking for. Some source code, which allows you to tweak physics at core, are open source. Google MRST, ADGPRS, OPENFOAM, etc. I would expect some tweaks to be already available in those codes. This will require some investment on your side into the selected code. It is probably advisable that you select one that is used by people around you.

A very crude alternative is to tweak an industrial grade simulator with restarts or use some of the options that exist in some of those "a little bit far". But this is very crude and messy, I would not recommend that.

Hi. 

I like to recommend to take a look at Majid Hassanizadeh's Publications. Here is an example of him:

Also, there is an other paper, entitled:

I think you can find your answer there. 

Please let me know  if you need more information. 

Hope it helps,

regards,

Alireza.

Use. SPIGOT software

Hi, 

It 's been long ago that  I was simulating multiphase flow.

I hope the following question are helpful for you. 

What is the detail information in the Out-File of the solver?

(A additional information could be a backup-file after the 5th and the 8th timestep. )

Is the transient timestep in a realistic scale of the massflow through the control volume?

One per cent of the pass-through time of the steam ought to be a reference value. The control volume oughts to be very big for a timestep of 1h. Typical timesteps are under 1 second.

Do you use the thermal energy equation or the total energy equation for solving the problem?

What happen if you only simulated the inlet condition with phase model without the reservoir? 

Is the absolute pressure near the inlet physical (wet steam)?

Have you define a reference absolute pressure in your control volume?

Good luck!

You may also look at the attached SPE Distinguished Lecture about the upgridding and upscalling.

Yes. But it may not be so simple and straight forward. A good understanding of rock geomechanical properties, exisitng stresses and their orientation including their impact on existing faults and fractures could be required.  Available fomration integirty test  and their associated issues ?

You can use Predici software for emulsion and liquid -liquid simulation.

My best wishes

There should be good agreement between log porosity and core porosity.  Contrary to this some times it happens that to regress log porosity with log permeability is inconclusive/strange.  Same could be happen when try to regress core permeability with core porosity.  In the absence of any correlation we can use the nearby formation information (existing wells from the same formations).  Last but not least it is always required to correct core perm to the log the log scale bcoz of the change of environment (from small scale to big scale) as well as insitu conditions.

I would suggest you read the experimetal work on "Positive imbibition capillary pressure curves using the centrifuge technique" by M. FLEURY, G. RINGOT and P. POULAIN. I am certainly sure with that, all the issues raised above wil be solved. 

As Khaldoon mentioned, 1) PVTi is part of SLB ECLIPSE suite of software but different from what most people refer to when talking of ECLIPSE. 2) There is no other way than relying both for input and output from PVTi onto files (typically txt). I am a bit rusty with PVTI but I am not sure that you can actually automate what you want to do in PVTi to build an automated optimization loop where MATLAB governs the overall process. From memory PVTi is very much scroll and click not macro geared.

Thank you for yours answer

Actually, that is not really correct because you argue with a pore scale picture while the reservoir simulation uses Darcy scale concepts.

You need to include capillary pressure into the simulation in all cases where capillary forces are comparable or larger than the viscous forces. That is the case for instance in the a transition zone, because there the saturation is strongly influenced by capillary pressure. In other cases capillary pressure might not be important, i.e. In many waterfloods it may not be needed. If in doubt, include it (as a guess value, using reasonable parameters) in your simulation and see if it makes any difference. But you have to make sure to use the right imbibition or drainage pc.

thank you Dr. Nawbahar for suggested of articles

Perhaps the main issue with standard reservoir simulation (Darcy-law based, single or dual porosity/permeability) for unconventional reservoirs is how do you specify matrix permeability. Typically the permeability of unconventional reservoirs is extremely low (of the order of 10-100 nD) which is below the operating range of standard laboratory equipment (which is approx. 0.1 mD). The specialized equipment which is used to measure nano-range permeability often provides ambiguous results as currently there are no standards available. Besides, there is big uncertainty related to the way you initialize the near wellbore area (what is the fracture permeability, which transfer functions should be used etc). It is not uncommon to see that your model will not reproduce the actual production data, thus forcing you to adjust the physical parameters with little to no justification.

Dear Dr. Youngmin Lee,

Thanks a lot for your response and your valuable information.

Thanks

Maleaha Samin

Dear Jai,

Thanks for you answer.

There are lots of researchers who have done coupling Eclipse reservoir simulator with Abaqus software. For example, there is an Abaqus plug-in which has been exclusively developed by Dassault Systèmes (Simulia Company) for linking Eclipse with Abaqus. In addition, many authors have developed C++, Python, FORTRAN, etc. codes that automatically couple Eclipse with Abaqus.

Most of them have used Partial Coupling scheme. First, the finite difference reservoir model runs within Eclipse software and pore pressure changes due to fluid injection/production are calculated at each user defined time step. At each user defined time step, the calculated pore pressures  are imported to Abaqus as input. Then, Abaqus calculates the stresses, strains, and displacements due to the pore pressure changes and updates reservoir parameters such as porosity, permeability and rock compressibility in the same time step. The updated reservoir porosity and permeability or compressibility are sent (exported) back to Eclipse reservoir model. Then, the reservoir finite difference model runs again within Eclipse using the new update porosities/permeabilities/ or compressibilites and calculates new pore pressure/saturations. The new calculated pore pressure are exported to Abaqus to recalculate stresses, strains, and displacements and to update reservoir porosity, permeability, and/or rock comressibility. This iterative cycle (data updating and exchanging between Eclipse and Abaqus) is repeated until reaching convergence at each time step (convergence means the difference between two iterations pore pressures is too low). 

If convergence is achieved, the same iterative operation is implemented for the next user defined time step.

This is a type of coupling between Eclipse (finite difference) and Abaqus (finite element).

However, I also need something to bring (import) Eclipse reservoir model geometry to Abaqus. I think I need to convert finite difference grids to finite element mesh. How can I use/import the same reservoir model geometry of Eclipse in Abaqus environment?

My opinion is No. I agree with Ann. And the same conclusion is true for every model.

If you are testing some feature of a model (let's say production), accuracy in the prediction of that feature does not guarantee accuracy in all it's other features.

Probably the best method for you to test your geostatistical reservoir model (seismic inversion, Acoustic Impedance, Elastic Impedance, Porosity, Lithologies and so on...) is to test it against blind data (because that data is real).

Please refer to following book to estimate such heavy products:

1) PVT and Phase Behavior of Petroleum Reservoir Fluids

Ali Danesh

2) Characterization and Properties of Petroleum Fractions

M.R. Riazi

As you stated above, properties estimation of these heavy products always has some time huge deviation from reality. An alternative procedure is to refer to experimental procedures by ASTM for example to estimate some transport and thermodynamic properties:

Volume 05.01 to 05.06 and 06.01 to 06.04

They use microseismic data to have a general idea of where the fracture goes. The idea is, during fracturing, fast slip might occur on both propagating fracture, or in the natural fractures in the medium, this can cause seismicity which can be detected by the detectors.

This is one of the most debated and open ended questions simulation guys are faced with. These are a few heuristics that might help you in the process provided you have lab PVT data to calibrate your model with:

Volumetric data - Accentric factors, omega values, critical conditions (not recommended)

Viscosity data - Z (critical) and V (critical)

Kindly note these are just a few observations from literature and can serve as a good starting point should you need one.

Selecting the coarser grid blocks (lower number of cells) in the compositional modeling through CO2 EOR flooding cause the larger effect of numerical dispersion. The presence of numerical dispersion has the effect of smearing out the recovery curves and make the determination of a break point more difficult / inaccurate. Hence, the task of determining the MMP becomes a balance between acceptable accuracy and acceptable CPU time consumption.  Further, as explained by Zick and Stalkup, the MMP can always be found by compositional simulation, though fine grids and repeated simulations at multiple pressures and certainly a significant computation times will be required.

You simply changed the composition of the stream. Therefore, your results changed. Please let me know if you want me to write the equations. You  increased lighter components and decreased heavier ones by increasing gas to oil ratio and thus everything changed. 

Hope this helps and all the best,

These incorporating parameters have significant effect on the recovery factor. Specially, Permeability that influences on the cut-off calculation

Dear Khaldoon,

Well, selecting the proper grid from the many available choices can be quite a challenge. Till date, I am not sure if there is any research paper on specific equations to choose proper grid is available that you looking for. Yet, this is a very interesting query you raised --it is a nice research problem. Probably we need to define self organized algorithm which will choose optimum grid as any optimization method does.

Please let me know if you identify any approach to solve this.

Thanks

I have used both CMG and Eclipse. I think CMG is better for a first time user (more user friendly). However, once you know the basics and the concepts you should start using Eclipse due to the fact that Petrel and Eclipse are more popular in industry as a connected suit. If you want to learn the coding of Eclipse then you should use the manual and/or skeleton codes available. My best suggestion to you is to read the technical manuals of each software. Learning how to do something is important but what is even more important is learning how the software is doing it to understand its limitations. Start slow with understanding basic Material Balance, conservation of mass and conservation of momentum. Once you know what the software does and how it is doing it, you should be able to use any dynamic simulator by just learning the coding (simple practice with time). Good luck and do not hesitate to ask/contact if you need any help. 

Reza,

Please try this keyword: WRFTPLT. It exports both RFT and PLT data. Search in the manual what it exports and what you need to specify. As far as I remember it is in the schedule section.

Not sure about PVTi Eclipse, but CMG- WinProp should be much easier.

It would be helpfull if you could be more specific on the problem that needs to be solved. The information you are providing is of general nature. Another thing that would be helpfull would be to provide some stratological description of the formation and its spatial relation to the reservoir.

If for example the problem at stake is the high salinity caused by disolution of this formation and the formation "touches" extensive areas of the reservoir at high water levels, one alternative could be to operate the reservoir (lower its level) so as to minimize contact area; that is of course if the economical losses from energy production are deemed to be acceptable.  But again, this is only a wild guess without much thought, since there is no precise information. If you are more clear about your request and provide more information, you are likely to get better answers.

If you're considering the stability of the IMPES method the classic reservoir simulation textbook by Aziz and Settari is well worth looking at.  The relationship required between del X and del t for stability is thoroughly presented there.

IMPES method has two source of stability limitations: primary variables and transmissibility.

During IMPES method, capillary pressure term is treated explicitly which cause stability issue depending on the Pc-curve's magnitude. Fluid comprerssibility changes the stability limit. Nevertheless, it is common to consider the stability limit under the incompressible assumption for practical purposes.

Besides, explicit treatment of transmissibility term leads to strong non-linearity. The well known stability limit due to this is that flood front can only advance one grid block per iteration. Otherwise, the total throughput of any grid block per iteration must be less that its pore volume.

Ref: Aziz, K., and Settari, A. 1979, Petroleum Reservoir Simulation. London, UK: Applied Science Publisher

           

The word "heterogeneity" means that the body is a mixture of different materials. "Anisotropy" describes a body that is made up of one single material that has different properties in perpendicularly different directions, that is, wave velocity is different depending on the direction by which the wave transverses the body. Perhaps it would help to look up the definition in a dictionary.

It is always desirable to perform relative permeability experiments  with real fluids at reservoir conditions. However, if you only have model fluids you can estimate the effect of the gas on relperm by including interfacial tensions or Capillary number if the wettability does not change.

You can get more details from:

Drainage and imbibition relative permeabilities at near miscible conditions.

Journal of Petroleum Science and Engineering

Volume 53, Issues 3–4, September 2006, Pages 239–253

 

It depends on your goal. <br />Your search for distributied model or a lamp on? or event base or continuous model?

Your goal,  available data, scenario which you are considering change in model, determine the model<br />

Time series (weather, rainfall etc..) modeling and prediction using soft computing techniques (Neural network, Evolutionary Computing, Support Vector machine etc...) is a good area and has practical applications.

J. Chen, J.W. Hopmans*, M.E. Grismer, 1999. Parameter estimation of two-fluid capillary pressure–saturation and permeability functions. Advances in Water Resources Vol. 22, No. 5, pp. 479–493.