[转] 水平井数据的使用讨论（英）

Joseph

Using log data from horizontal wells in reservoir modeling

David ContrerasGeomodeller at Schlumberger Information Solutions

Dear colleagues would like to discuss the best practice in order to use log data from horizontal wells into reservoir modeling. Most of our current model updates are related with this kind of data that most of the time is quite tricky to make it work on the modeling
Detailed QC is perform during log scaling (bias, cell selection, grid orientation) in order to capture the data properly, additionally we are starting to analyze whether using the same vertical variogram range or increasing this distance in order to let the horizontal data to propagate further.

Thanks in advance for your comments
DavidC

Hadi Alikhani

Development Geologist

Dear David,
I think, you should ignore horizontall wells for generation of vertical variograms, if you have enough wells (only for vertical variograms and only for zones which horizontal legs have crossed them).
If the wells are not enough to ignore, you should take the band width so that cover the whole lateral distance of the leg, but less than the minimum wells distance, too.
actually, you must use of input data, not up-scaled data for generation of vertical variograms especially, in this situation.

Ashkan Asadi

Reservoir Geologist/Lead Geologist (NIOC)

David, how many horizontal wells are there in the field you are dealing with!? What about the number if verticals!?
What I can generally suggest is that: first of all, the horizontal wells should only be used for determination of horizontal ranges. Please notice that the amount of horizontal range might not necessarily increase after using horizontal wells. In fact, horizontal wells are very good indicators for examination of the previous ranges you had defined for your model. If the new wells show some extra continuity then you should increase the earlier ranges but if the horizontal wells confirms the original ranges, then just let them go as they were. Secondly, you may produce horizontal variogram map by the use of horizontal wells and compare it with earlier ranges. In agreement with Hadi, I also recommend to use original well logs instead of up-scaled ones. All the best. Ashkan

David Contreras

Geomodeller at Schlumberger Information Solutions

Hadi, Ashkan thanks for your comments...

I did a couple of exercises: one using only verticals (211 wells) another using 32 horizontals for a total of 242 wells.

The ranges do not change dramatically with the horizontal data, the main difference when using the horizontal data is that apparent nugget tends to disappear but the variograms becomes quite noisy.

The main issue is that my vertical range (from verticals only) is about 1-1.5 m and horizontal resolution is about 50 meters, so when dealing with PHIE and PERM, the averaging used during upscale incorporate odd values for horizontals (even when BIAS filter is applied).

On the other hand whe working with verticals only the data analysis and variograming is quite better but the simulated results do not honor the horizontal data in most of the horizontal paths (ranging from 500-900 m).

The best result I have achieved so far is using all the wells (verticals & horizontals) but performing up-scaling with BIAS (form rock-types) and through cell method (in PETREL), with that setting the results are quite better than using only verticals.

Thanks for the comments
DavidC

Glenn Krum

Geologic Modeling Consultant

You can use horizontal wells that have not been scaled up for horizontal variograms if you have to, but frequently they cross stratigraphic layers so are best not used unless you can compensate. Their best use in generally for facies (something with less variability within a cell rather than fine scale data like porosity and permeability. Using facies from horizontals allows you to populate properties by facies and better match well data/performance. If you do not use facies from the horizontal and facies are an important control in your reservoir then there may be a mismatch between the model and the horizontal well performance due to the statistical variation between vertical wells that is uninformed by the horizontals.

Hadi Alikhani

Development Geologist

Dear David,
Actually, horizontal wells are always very problematic to be used in the property modeling. The most of the problems are due to up-scaling step and because of horizontal size of the grid-cells.
Are you sure that, the horizontal part of the legs, were correctly situated in the real place of the grid? for example in the real place below or above the top-surface? It is very critical issue and must be considered carefully.
Also I suggest you to check the anisotropy directions by surface variogram, removing trends and analyzing distribution of the data.
To my mind, the number of vertical wells are enough to do the variography and petrel is a very powerful software in this way.

Good luck.

David Contreras

Geomodeller at Schlumberger Information Solutions

Dear colleagues I really appreciate your comments

Totally agree with Glenn about considering only the discrete logs for up-scaling in the horizontal legs, and then the resulting propagation for continuous properties must fit trends and honor well data
As I mentioned the data is keeping directions of variability and ranges are quite the same whereas horizontal data is used or not.
Talking about the structural position of the horizontal paths the team pays lot of attention to that either for the landing portion of the boreholes and also for the final meters of the trajectory where few of the wells exits a given stratigraphic level.
I would like to summarize this helpful discussion with the following thumb rules:

1) Log data form horizontal wells is helpful to confirm/refine anisotropy and horizontal ranges for variograms

2) Discrete logs (facies or rock-types) in horizontal wells can be used for upscaling in order to honor data for the horizontal legs

3) Continuous data for horizontal wells should not be used for upscaling in property modeling but data from the non-horizontal portion might be includes in the upscaling.

Again would like to thank you all for the participation and more suggestions will be considered

Cheers
DavidC

Ashkan Asadi

Reservoir Geologist/Lead Geologist (NIOC)

David, you have concluded very well but just one point, the main reason behind better model performance after using facies-biased upscaling in horizontal wells is that, in SGS/SIS process the well data is completely honored and adjusted while modeling and hence the simulation model better matches with geological input. In fact, the horizontal wells do not influence the range however they influence better data population in those cells exposed to horizontal wells.
It was a very interesting discussion followed with a lesson learning. Regards.

Alexey Romanov

Geologist and Partner at Sproule Associates

Hello all,

I would like to add my two cents to the discussion:

1. Horizontal wells are drilled intentionally in the best parts of the reservoir to maximize production per well. This results in overoptimistic reservoir statistics, when horizontal and vertical wells are upscaled and analyzed together, just because the best reservoir is sampled more often by the horizontal wells.
So, the representative reservoir mean parameter should be estimated after declustering and on the vertical wells only.

2. However, the horizontal wells provide additional "known" points for a geostatistical algorithm and they could be used as input points to control surrounding cells.

3. Another good use of horizontal wells is that they provide additional control on reservoir structure, layer dipping etc.

4. I'm not really sure, that horizontal wells add any value for variogram calculation, too.
Firstly, they are useless for vertical variograms;
Secondly, directionality of the wells may introduce an impression that correlation in a well direction is higher, just because we have more data points lined in this direction. Another argument could be that log quality from horizontal wells (different tools, another resolution, different influence of ambient rocks on readings etc) is different from the conventional logs run in vertical wells. And mixing the data of different origin in one analysis (variography) could be misleading.

So, my suggestions are to use the horizontals for structural control and as additional seed points for simulation, but be extremely cautious with any conclusions on reservoir statistics and/or continuity coming from any analysis involving the horizontal wells.

Cheers,
Alexey

Joseph

Dian Kurniawan
Sr. Reservoir Geologist at Eni Indonesia

Hi all,

I'm totally agree with Alexey.
I just want to share my experience using horizontal well data in reservoir modeling.
Horizontal wells data give "bias" in variogram estimation and data distribution. Alexey already mentioned that Horizontal well (commonly) are drilled intentionally in the best parts of the reservoir, there is a risk when we used horizontal well data in geostatistical estimation ---> the result of property distribution would be over-estimating. So, I didn't use horizontal well data in variogram estimation. Here is my workflow :
1. Do up-scaled property log from vertical wells and wells that have inclination <50 deg (it's just rule of thumb that I applied as a thresh hold so that I can eliminate "bias" from horizontal wells data). If I'm not mistaken in Jewel, we can set a thresh hold for the intersection angle between wellbore and model

2. Do variogram estimation by using those up-scaled data until I get all parameters (vertical/horizontal range in major/minor direction, nugget etc.)

3. Up-scaled property from all wells (including horizontal wells)

4. Use results from variogram analysis (step.2) for distributing the property

Please always do QC step by using blind test method. If the comparison between blind test vs actual data is close, I think the model is OK
Anyway, Horizontal wells data are useful to QC lateral distribution & continuity of the reservoir.

(cmiiw)

Thanks & regards,
Dian

Hadi Alikhani
Development Geologist

Hi experts,
I exactly agree with Dian.
Also I think it's not so related to being discrete or continues data. please correct me, if i'm mistaken,
As you know, the stationary condition is a vital rule in geostatistics. When we use the horizontal data, we will have dense values in some specific lateral directions and it will disarrange dispersion of the data and statistical population consequently. So our variography will be affected by the diffrent sampling distribution.
But I've had an exprience about this problem and I found a solution for it by petrel upscaling options. If you have to use all wells in variography same as me, It's not bad to see my work follow on this link, it may be helpful.

https://skydrive.live.com/redir? ... ithint=file%2c.pptx

Best Regards,
Hadi Alikhan

David Contreras
Geomodeller at Schlumberger Information Solutions

Dear all, such a good discussion and would like to thank for highly detailed comments and approaches.

I totally agree with Hadi in his workflow, especially for the 'trough cell" selection of cells when performing scale-up-well-logs. It is strongly recommended to use bias filtering (with facies) as suggested by Ashkan in order to assign the best approximation of a given property per cell.

With this method there is substantial difference in the number of scaled cells as the through cell method gives fewer cells than the remaining options

Do agree that statistics shall be over-estimated by horizontal data so variogram analysis without horizontal wells is quite fair and non-trendy due to data distribution.
So would like to postulate an improved workflow following Dian’s guide lines:

1. Do up-scaled property log (using BIAS) from vertical wells and wells that have inclination smaller than the tolerance defined by user.

2. Do variogram estimation by using those up-scaled data until get all parameters (vertical/horizontal range in major/minor direction, nugget etc.). Mean and variance would not be affected by data from horizontal legs

3. Up-scaled property from all wells (including horizontal wells) using through cell selection in order to capture fewer values in the horizontal path

4. Use results from variogram analysis (step.2) for distributing the property. When using SGS method it is strongly recommended to apply Ordinary krigging in order to get a grid of over-estimated means and smaller variance around the legs. Additionally would like to point out that distribution of discrete data (facies-rock types) shall be performed before running continuous data distribution.

With this procedure we are accepting to use horizontal data in reservoir modeling which is the main point of the discussion. However the use of this kind of data must considered the workflow mentioned above

Once again let me express my gratitude for your participation in this brainstorming

Regards
DavidC

Alexey Romanov
Geologist and Partner at Sproule Associates

Hello David,

Could you,please elaborate more on reasoning to use ordinary kriging for SGS?
Simple kriging is used in theory for SGS.

Cheers,
Alexey

David Contreras
Geomodeller at Schlumberger Information Solutions

Dear Alexey here is my thought:
There are two ways for running Sequential Gaussian Simulation (SGS), the classic is with Simple Kriging that involves a global mean during kriging, ant the second one is with Ordinary kriging that uses a local mean. When dealing with highly heterogeneous data, the ordinary kriging provides better results as the average distributed value varies according with the surrounding data.
Being said that, you can minimize the propagation of over-estimated data from horizontals because these extreme values do not participate in the global mean calculation.

Detailed information about the calculation please check --- http://www.statios.com/Training/ --- there is a section called ---8. Gaussian Simulation for Porosity modeling--- with extra details about it.
Let’s keep going with the discussion in order to have a clear idea for the suggested workflow.

Regards
DavidC

Joseph

Glenn Krum
Geologic Modeling Consultant

Alexey: Ordinary kriging requires dense well data, several well data points within the range of of the variogram. If your data are not this dense then the default of simple kriging should be your choice. The advantage of ordinary kriging is an estimate for each cell defined by the local data, it automatically produces variations in the estimated values at each cell to honor the local mean while simple kriging falls back on the global mean, everywhere but particularly away from data control. Modelers can add local variation by using trends or a LVM (locally varying mean).

Alexey Romanov
Geologist and Partner at Sproule Associates

David/ Glenn,

Thank you for your comments. My main concern is that in case of SGS with ordinary kriging there has to be significant data control.
In other cases risk of spreading data values beyond their range of influence is high.
Therefore I would not use the phrase "strongly recommend to use OK with SGS for cases of modeling with horizontal wells".
SGS with OK may work in some cases better then SGS with SK, but in my opinion it cannot be a general rule.

Cheers and Merry Christmas,
Alexey

Hadi Alikhani
Development Geologist

Hi experts,
Good comments.!
Also, ordinary kriging is so sensetive to the local changes of values, mathematically.
So I think that, heterogeneity of the property must be considered firstly, because its so important in making decision about type of the kriging method. In other word, if we have local heterogeneity we must use ordinary kriging, otherwise its not different.
Also OK is so sensetive to the out-layers and the range of values.
Best wishes and Merry Christmas,
H.Alikhani.

Oliver Torres
Director at GEO MODELS LTD

Hi all,

Very interesting thread.

1) In order to obtain a representative histogram horizontal wells should be filtered out at least you have a declustering tool which unfortunately Petrel does not have it.

2) For variogram calculation you don't need to filter out horizontal variograms. The whole idea of doing a variography study is to calculate variograms a different vectors (then horizontal ones) with small search cones. Vertical variograms of the whole data set will not be influenced.

3) I would always prefer to use ordinary kriging because it will use a local mean by wells around the target. I still dont know why SLB has set up the simple kriging as default.

Kind regards

Oliver Torres
Geostatistician

Joseph

这个讨论是从LinkedIn网站的Geomodeller讨论组里粘过来的，原文地址：

https://www.linkedin.com/groupIt ... ms_see_more-0-b-ttl

howardfrank

感谢分享！

zhj_0074

感谢分享！

kurtli

类似这样的技术贴， 从概念到细节 将书本上的概念具体化 在国内 俺只在郝大神的帖子里看到过。多谢大神分享。

qingsong88

帖子出来了，谢谢郝老师

petroyb

谢谢郝老师，详细阅读学习下了！！