Genetic algorithm application for matching ordinary black oil PVT data

Petroleum Science, Jul 2012

In the study of reservoirs, it is vital that we have a realistic physical model of the reservoir fluid that accurately describes the hydrocarbon system and its properties. The available equations of state (EOS) to model the fluid phase behavior have some inherent deficiencies that may cause erroneous predictions for real reservoir fluids, so these models should be tuned against experimental data by adjusting some parameters. Since there are many matching parameters, tuning the EOS against experimental data is a tedious and difficult work. In this study, a genetic algorithm as an optimization technique is used to solve this regression problem. This study presents a new method that uses a specially designed genetic algorithm to search for suitable regression parameters to match the EOS against measured data. The proposed method has been tested on three real black oil samples. The results show the surprising performance of the developed genetic algorithm to match the experimental data of the selected fluid samples. The main advantage of the used method is its high speed in finding a solution. Also, finding more than one solution, working automatically, confining the role of experts to the last stage, reducing costs and having the possibility of evaluating the different situations are the other advantages of this method to match ordinary black oil PVT data and makes it an ideal method to implement as an automatic EOS tuning algorithm for black oils.

A PDF file should load here. If you do not see its contents the file may be temporarily unavailable at the journal website or you do not have a PDF plug-in installed and enabled in your browser.

Alternatively, you can download the file locally and open with any standalone PDF reader:

https://link.springer.com/content/pdf/10.1007%2Fs12182-012-0200-2.pdf

Genetic algorithm application for matching ordinary black oil PVT data

Pet.Sci. Genetic algorithm application for matching ordinary black oil PVT data Mohammad Taghizadeh Sarvestani Behnam Sedaee Sola Fariborz Rashidi designed genetic algorithm to search for suitable regression parameters to match the EOS against measured data. The proposed method has been tested on three real black oil samples. The results show the EOS tuning algorithm for black oils. - 1 Introduction upon oil and gas samples. Experimental data are obtained and used in combination with an equation of state model to conditions. substances and carbon groups. The carbon groups are not the critical properties of the carbon groups required for EOS deficiencies is to calibrate, or tune the EOS models against experimental data. There are no well defined rules for how to do regression of an equation of state model to match to (1986) contains an appendix on the choice, selection and equation of state. state parameters. This tuned model is then can regarded as pred Wj j X i j j (1) exp represent predicted and experimental W is the weighting factor, Ndata expresses Xi classes of search techniques like calculus-based techniques, algorithms. Strong features of genetic algorithm such as its based on a stochastic-directed trend with roots on ideas from et al, 1995), and aircraft design (Parnee and Watson, 1999). production data on a structural model. matching PVT data for three real black oil fluid samples is 2 Description of the method the regression parameters in detail. against the experimental data, matching parameters of weighting factors for different properties of the fluid which pseudo-components if splitting of the plus fraction is needed, Then the software uses the selected parameters to perform selects another set of items for the next run. These trial and is found. the EOS model, assigns weighting factors, and determines interaction coefficients and matching parameters of the the different cases for the number of pseudo-components three different situations regarding the number of pseudocomponents were tested. These were: no splitting, 2 pseudoweighting factors, the general rules were applied. For example, the highest weighting factor was assigned to the are fed into the commercial PVT software, which has been coupled with the program, as the input data. The commercial and presents the best solution. The regression method used in the PVT software uses the Newton numerical method to find the minimum of a residual function that is defined calculated data matrix. The residual function depends on a predetermined number of iterations for some special cases the program and the new fluid model replaces the original against experimental data. Furthermore, sometimes the tuning the program compares these 3 solutions and selects the best iterations since the weighting factors do not change through degrades it. Fig. 1 shows three different cases extracted from SM0.16 R 0.19 0.18 0.17 0.15 0.14 0.13 1 Fig. 1 2 Iteration 3 a b c parameters in this case. This process is iterated for the 30 different cases and at The phase diagram of the final model should represent The complete process is shown in Fig. 2. 3 Genetic algorithms and genetic for the design and implementation of robust exist. The genetic algorithms start with an initial population the current population of the solution is known as the parent the genetic processes occurring in the nature, i.e. selection, and their genetic information is recombined and modified to generate the new population known as offspring. The offspring are inserted into the population, replacing the parent population and producing a new generation. The genetic and representing them in a genetic format, the selection of new solutions. No splitting 2 pseudo-component 3 pseudo-component 3-PR 3-SRK chromosome for the EOS parameters is a two-dimensional Produce the next population (Offspring) Solution 1 Solution 2 Solution 3 . . . Solution 30 A certain number of population A suitable individual found Selection: Select the good parents Mutation Crossover NO YES Condition Result Selection of the good models by user 3.1 Initialization The information that is to be held within the genome is spite of the fact that the general structure for genome is a each group being allocated to a separate chromosome. The Z regression parameters in the chromosome of the EOS and example, the plus fraction row in the matrix has the highest Z the parent population. 3.2 Selection 0 0 0 0 1 1 C1 C6 0 0 0 0 1 1 0 0 C7+ 0 0 0 0 0 1 Lohrenz-Bray-Clark Pedersen Aasberg-petersen this population are taken and used to form a new population population will be better than the old one. Solutions which are then selected to form new solutions are selected according where N Wi 1 1 xi (2) the search process. Xi represents population (S). (3) i M This method is the simplest proportionate selection chromosome. Fig. 6 shows the principle of this selection method. Fix pointer 1 5 6 The wheel is rotated and the chromosome that stops in sum is greater than the random number, stop the summation This loop is iterated while the required numbers of chromosomes are selected. 3.3 Crossover and mutation operators The selected parents should produce the next population x-coordinates x and y-coordinates of these numbers are dimensional chromosome. Fig. 7 shows different situations x and y-coordinates of the random numbers. A=(a1, a2) B=(b1, b2) A=(a1, a2) B=(a1, a2) A=(a1, a2) B=(b1, b2) A=(a1, a2) B=(a1, a2) this chromosome should be selected that shows the method of nor select none. So, this operator will not perform on these and flip (0 to 1, or 1 to 0) the gene in the two-dimensional chromosomes. number is less than 0.33, the program will flip the first block and if the number is between 0.33 and 0.67, flip the mutation is 1, no change will be occurred. This is because change with respect to its generated random number. For C and the mutation M C and M, When the offspring population is produced, we compare 4 Sample data are checked. from two experiments performed on the fluids: constant at saturation pressure. 5 Results and discussion answers are produced for each sample. For black oil-1, six nine answers were acceptable as engineering aspects. Fig. 8 to Fig. 10 show the results of the tuning of EOS against calculated for different properties. Table 7 shows the results equation was used: algorithm is its high speed in finding the solution. While in the other answers could be related to weighting factors. 3.5 3.0 manual tuning these factors can change through the process for each case, we would expect different results and tuned Measured Calculated Measured Calculated 58 56 54 3 ft/ 52 b I, 50 y itsn 48 e ild 46 O 44 42 40 1.35 1.30 B /TS1.25 B ,R1.20 F V ilF 1.15 O 1.10 1.05 1.00 models for each case if we re-run the program. is run 10 times for black oil-1. Fig. 11 shows the number of acceptable solutions as engineering aspects for 10 different factors change in each run, the following equation is used in * * 1 (5) Pressure, psia Measured Calculated 2000 3000 Pressure, psia Measured Calculated 4000 5000 3.0 2.5 1.00 0.98 a suitable method in tuning EOS against PVT experimental 0 0.17 0.16 0.15 0.14 *S0.13 RM0.12 0.09 1.5 1.4 % , r 1.3 ro r e 1.2 g e a r e 1.1 v A 1.0 3 4 5 3 4 5 6 Conclusions The results of this study show that the developed genetic and time consuming operations of tuning of EOS against property graphs show the successful tuning of EOS against measured data; furthermore, the average error values are below 2 percent for all the cases and prove that the GA we cannot argue that the method presented in this study is the The strong non-linearity of the EOS tuning process makes classical deterministic optimization methods inefficient approach would be to use heuristic type methods like continuous and discontinuous variables, changing several variables simultaneously and the ability of this method to work with different data structures in the same time, cause this optimization method to be a good choice to solve the The main advantage of the method is its high speed difficult work even for an experienced reservoir engineer and often needs a long time to find just one tuned model, the proposed method can find more than one solution in a of experts to the last stage, reducing costs and having the possibility of evaluating the different situations are the other advantages of this method to match PVT data and makes it an ideal method to implement as an automatic EOS tuning Comparing the RMS values for different iterations shows that modifying the matching parameters indiscriminately does not develop the model necessarily and may results in the Since the GA is a stochastic algorithm, different results are achieved for different runs of the program and the proposed Introductory Analysis with Applications to Biology, Control, and


This is a preview of a remote PDF: https://link.springer.com/content/pdf/10.1007%2Fs12182-012-0200-2.pdf

Mohammad Taghizadeh Sarvestani, Behnam Sedaee Sola, Fariborz Rashidi. Genetic algorithm application for matching ordinary black oil PVT data, Petroleum Science, 2012, 199-211, DOI: 10.1007/s12182-012-0200-2