The use of experimental design (Design of Experiments) | VV's Blog
In this post I want to talk about the application of the method of experimental ranstad self service design (design of experiments) for the analysis of uncertainties. In most cases, when the result is dependent on several uncertain parameters can not be obtained within a few seconds after the change of the input data, I prefer to use this method. This saves a lot of time and still get very good results.
First, the background. A month ago, it was necessary to urgently insert a new not previously approved in the well drilling program. At the moment, according to our model well looks very attractive, furthermore reserves and debited to the start better than any other wells. Wherein the well is located in a part in which there are still high enough porosity uncertainty, NTG, permeability fractures, the presence or absence of channels vysokoprovodimyh vertical anisotropy (Kv / Kh) and so on. In general, there are many doubts on account of the fact that this is the best well in the project. The model on which to base our calculations, more precisely, three versions of the model - pessimistic, most likely and optimistic (Low, Mid and High) - zamatcheny of history in the light of seismic ranstad self service 4D. And in general, across the field, these model variants ranstad self service roughly correspond reserves ranstad self service with P90, P50 and P10. Detailed probabilistic modeling was carried out a few years ago, and after that there was a very serious edits. But it turned out that in the zone of this well run by the properties was minimal, which in principle does not contradict ranstad self service anything. As a result, cumulative production of this well is not very different in all three options ranstad self service (here we must note that we are talking about incremental spoils ranstad self service e - I think that this must be a separate post napisat s somehow). Minimum run-on options, was untenable, so it was decided to carry out a detailed uncertainty analysis, but only to the portion of the formation, which will be located well, not to destroy the adaptation model. Fortunately, this part of the formation practically did not interact with the developed part of the field, so any manipulation with properties not greatly affect ranstad self service the operation of existing wells.
In the traditional Monte Carlo simulation, to get at least some idea of the distribution ranstad self service function stocks well (in this case we are talking about the accumulated reserves of the well), had had to spend more than one hundred calculations. And considering that in our case, the result must be obtained by conducting the calculations in the simulator, spending 2 hours on a version ranstad self service with a maximum of 5 parallel calculations on a cluster, then it would take a long time. Spend time working on these calculations is extremely impractical.
In this case, the experimental design is a great way to significantly reduce the number of calculations (in our case - the number of model runs), and that is also important number of required input data.
Without going too much into detail and put it in simple terms, ranstad self service I would say that the essence of the method is as follows (here we go into plain language for the oil industry, with examples) - for each of the uncertain parameters (such as, for example - porosity, NTG, water saturation, permeability level BHK end points permeabilities, etc.) are given possible boundaries from minimum to maximum. Different combinations that give a certain value of the output function - STOIIP, recoverable reserves, etc. Also worth noting is that the boundaries of the minima and maxima as possible must be set wider to cover the entire range of possible ranstad self service values. That was covered with interval probability value from close to 0 to close to 100%. Ie in other words you need to sgenenirovat in addition to the average, two deterministic variant ranstad self service Low-Low and High-High.
If you omit the fact that the planning of the experiment, depending on the task used different ways to construct the table plans for our oil problems very well suited method of Plackett-Burman (PB). Read about it can be, for example here: http://www.itl.nist.gov/div898/handbook/pri/section3/pri335.htm. In this context, to assess the factors used in all N N + 1 starts.
The first row of the design matrix is given in the form (for example m = 11). Then each subsequent row of the matrix formed ranstad self service from the previous cyclic shift to the right. The last row of (m + 1) consists of (1). As a rule, this plan adds two lines with one and the central point - 0, which is also used to check the adequacy of the proxy.
But it also should be noted that pure Plackett-Burman, usually only used for the initial evaluation ranstad self service in the experiment to identify ranstad self service the most important ranstad self service parameters and UTS
In this post I want to talk about the application of the method of experimental ranstad self service design (design of experiments) for the analysis of uncertainties. In most cases, when the result is dependent on several uncertain parameters can not be obtained within a few seconds after the change of the input data, I prefer to use this method. This saves a lot of time and still get very good results.
First, the background. A month ago, it was necessary to urgently insert a new not previously approved in the well drilling program. At the moment, according to our model well looks very attractive, furthermore reserves and debited to the start better than any other wells. Wherein the well is located in a part in which there are still high enough porosity uncertainty, NTG, permeability fractures, the presence or absence of channels vysokoprovodimyh vertical anisotropy (Kv / Kh) and so on. In general, there are many doubts on account of the fact that this is the best well in the project. The model on which to base our calculations, more precisely, three versions of the model - pessimistic, most likely and optimistic (Low, Mid and High) - zamatcheny of history in the light of seismic ranstad self service 4D. And in general, across the field, these model variants ranstad self service roughly correspond reserves ranstad self service with P90, P50 and P10. Detailed probabilistic modeling was carried out a few years ago, and after that there was a very serious edits. But it turned out that in the zone of this well run by the properties was minimal, which in principle does not contradict ranstad self service anything. As a result, cumulative production of this well is not very different in all three options ranstad self service (here we must note that we are talking about incremental spoils ranstad self service e - I think that this must be a separate post napisat s somehow). Minimum run-on options, was untenable, so it was decided to carry out a detailed uncertainty analysis, but only to the portion of the formation, which will be located well, not to destroy the adaptation model. Fortunately, this part of the formation practically did not interact with the developed part of the field, so any manipulation with properties not greatly affect ranstad self service the operation of existing wells.
In the traditional Monte Carlo simulation, to get at least some idea of the distribution ranstad self service function stocks well (in this case we are talking about the accumulated reserves of the well), had had to spend more than one hundred calculations. And considering that in our case, the result must be obtained by conducting the calculations in the simulator, spending 2 hours on a version ranstad self service with a maximum of 5 parallel calculations on a cluster, then it would take a long time. Spend time working on these calculations is extremely impractical.
In this case, the experimental design is a great way to significantly reduce the number of calculations (in our case - the number of model runs), and that is also important number of required input data.
Without going too much into detail and put it in simple terms, ranstad self service I would say that the essence of the method is as follows (here we go into plain language for the oil industry, with examples) - for each of the uncertain parameters (such as, for example - porosity, NTG, water saturation, permeability level BHK end points permeabilities, etc.) are given possible boundaries from minimum to maximum. Different combinations that give a certain value of the output function - STOIIP, recoverable reserves, etc. Also worth noting is that the boundaries of the minima and maxima as possible must be set wider to cover the entire range of possible ranstad self service values. That was covered with interval probability value from close to 0 to close to 100%. Ie in other words you need to sgenenirovat in addition to the average, two deterministic variant ranstad self service Low-Low and High-High.
If you omit the fact that the planning of the experiment, depending on the task used different ways to construct the table plans for our oil problems very well suited method of Plackett-Burman (PB). Read about it can be, for example here: http://www.itl.nist.gov/div898/handbook/pri/section3/pri335.htm. In this context, to assess the factors used in all N N + 1 starts.
The first row of the design matrix is given in the form (for example m = 11). Then each subsequent row of the matrix formed ranstad self service from the previous cyclic shift to the right. The last row of (m + 1) consists of (1). As a rule, this plan adds two lines with one and the central point - 0, which is also used to check the adequacy of the proxy.
But it also should be noted that pure Plackett-Burman, usually only used for the initial evaluation ranstad self service in the experiment to identify ranstad self service the most important ranstad self service parameters and UTS
No comments:
Post a Comment