In this discussion
[link Point to another website Only the registered members can access]
in the Pet. Software section I got the idea that it would be a good idea to have a thread just about how you do this. We can not go too much into the details as we will never see the model someone is working on, but I think general ideas and suggestions might be sufficient to help people in this area.
For myself I can say I have done some but not enough to say I am an expert as in our business you need to do 10 different things at a time and can not often spend a lot of time in one area. That was my working environment, "Now your on project X doing y, then your on project Y doing s." Never in one very long.
Well here we go ....
In the discussion I mentioned above someone was asking about a program SLB has called SIMOPT. I had once looked at it in my spare time. Using gradient sensitivities the programs helps to find a better history match. That sounds easy in those few words but requires much more work and understanding of what is happening.
With this in mind I would suggest the following work available for free which explains a bit this process in another application.
The above comes from here which I recommend to readSensitivity-Based History Matching Algorithms and Streamline Methods
The prominence of sensitivity-based history matching algorithms can be largely attributed to the rapid convergence they exhibit. Because of the computational challenge posed by even the smallest of field-scale history-matching endeavors, it becomes imperative for the computation of sensitivity coefficients to be as efficient as practically possible. One of the distinguishing features of streamline-based history matching algorithms is their superior efficiency in computing sensitivity coefficients.1 It
is the rapid sensitivity computation and thus applicability of the streamline-based method achieved in two phase applications that motivates the extension to three-phase production data researched in this work. The efficacy of the approach in calculating sensitivities is a direct consequence of the nature of the streamline formulation for modeling the dynamics
of fluid flow. In the streamline domain, the flow and transport equations are decoupled with a resulting reduction of the solution of a three-dimensional problem to a series of one-dimensional problems.14 In chapter II, we discuss the streamline formulation for the forward problem, and the sensitivity formulation for the inverse problem is detailed in chapter III
[link Point to another website Only the registered members can access]
Another good article from someone doing the work for a client and what the client was complaining about is this one
[link Point to another website Only the registered members can access]
Some parameters are not important AT CERTAIN TIMES to the end result from a simulation. This is where sensitivities can tell you a lot (what to focus on).
Reply With Quote
Bookmarks