Vous êtes sur la page 1sur 16

SPE 38855 EOS Simulation and GRNN Modeling of the Behavior of Retrograde-Gas Condensate Reservoirs

Adel M. Elsharkawy and Salah G. Foda, Kuwait University, P.O. Box 5969, Safat 13060, Kuwait

Copyright 1997, Society of Petroleum Engineers, Inc. This paper was prepared for presentation at the Annual Meeting Conference and Exhibition held in San Antonio, Texas, USA, 5-8 October 1997. This paper was selected for presentation by an SPE Program Committee following review of information contained in an abstract submitted by the author(s). Contents of the paper, as presented, have not been reviewed by the Society of Petroleum Engineers and are subject to correction by the author(s). The material, as presented, does not necessarily reflect any position of the Society of Petroleum Engineers, its officers, or members. Papers presented at SPE meetings are subject to publication review by Editorial Committees of the Society of Petroleum Engineers. Electronic reproduction, distribution, or storage of any part of this paper for commercial purposes without the written consent of the Society of Petroleum Engineers is prohibited. Permission to reproduce in print is restricted to an abstract of not more than 300 words; illustrations may not be copied. The abstract must contain conspicuous acknowledgment of where and by whom the paper was presented. Write Librarian, SPE, P.O. Box 833836, Richardson, TX 75083-3836, U.S.A., fax 01-972-952-9435.

engineering calculations in case laboratory data is unavailable. The GRNN models save computer time involved in EOS calculations. The study also shows that once these models are properly trained they can be used to cut expenses of frequent sampling and laborious experimental CVD tests required for gas condensate reservoirs. Introduction A reliable prediction of the performance of gas condensate reservoir during pressure depletion is necessary in evaluating new fields and determining their reserve. The predicted performance is also used in material balance calculations and making economic decision regards planning future recovery processes. Such prediction of gas condensate performance can be obtained from experimentally measured data on constant volume depletion (CVD) test using bottom hole samples or surface recombined samples collected where the reservoir pressure is above the dew point pressure. In cases where CVD data is not available, equation of state and empirical correlation are used to predict the depletion performance of gas condensate reservoirs. Cubic equation of states (EOS) are used extensively in the petroleum industry for studying volumetric and phase behavior of petroleum reservoir fluids and to check the validity of the PVT data, Conrard and Gravier1 (1980), and measured saturation pressure, Raghavan and Jones2 (1996). EOS Models are also used for separator design, gas plant design, compositional simulations, and vertical performance calculations of gas condensates. A numerous number of equation of states (RedlichKwong3; 1941, Soave4; 1972, Peng-Robenson5; 1976, Zudkevitch-Joffe6; 1970, Martin7 ;1979, and Peneloux8 ; 1982) have been published in the literature to model reservoir fluid phase behavior in general and gas condensate in particular. Generally, the performance of the EOS is good for simple hydrocarbon systems, predominantly oil. However, it deteriorates for phase behavior modeling of complex hydrocarbons such as volatile oils and gas condensates specially in the retrograde region, Saker et al.9 (1991). An important step in any meaningful use of the EOS based compositional model is satisfactory agreement between

Abstract Currently, two approaches are being used to predict the changes in retrograde gas condensate composition and estimate the pressure depletion behavior of gas condensate reservoirs. The first approach uses the equation of states whereas the second uses empirical correlations. Equations of states (EOS) are poor predictive tools for complex hydrocarbon systems. The EOS needs adjustment against phase behavior data of reservoir fluid of known composition. The empirical correlation does not involve numerous numerical computations but their accuracy is limited. This study presents two general regression neural network (GRNN) models. The first model, GRNNM1 is developed to predict dew point pressure and gas compressibility at dew point using initial composition of numerous samples while the second model, GRNNM2, is developed to predict the changes in well stream effluent composition at any stages of pressure depletion. GRNNM2 can also be used to determine the initial reservoir fluid composition using dew point pressure, gas compressibility at dew point and reservoir temperature. These models are based on analysis of 142 sample of laboratory studies of constant volume depletion (CVD) for gas condensate systems forming a total of 1082 depletion stages. The data base represents a wide range of gas condensate systems obtained worldwide. The performance of the GRNN models has been compared to simulation results of the equation of state. The study shows that the proposed general regression neural network models are accurate, valid and reliable. These models can be used to forecast CVD data needed for many reservoir

ADEL M. ELSHARKAWY AND SALAH G. FODA

SPE 38855

laboratory PVT data and EOS results. Although, sampling gas condensate is a costly process and, sometimes, subject to errors in recording initial reservoir pressure and temperature, the laboratory data with all the errors it contains are needed for tunning of the EOS, Drohm and Goldthrope10 (1988). Knowledge of the phase behavior of the gas condensate systems is important to predict reservoir performance and future processing needs. Phase behavior calculations from EOS are desirable because experimental determination is laborious and expensive. The degree of confidence in equation of state results increases as EOS is tuned to match experimentally measured data or actual field data. Once the EOS is tested against actual data for a given fluid, it is generally assumed that it can be used to describe the phase behavior of that fluid at any condition, Al-Mahros and Tijoa11 (1987). A number of studies have been reported comparing EOS results and laboratory PVT data for various kinds of reservoir fluids (see references: Firoozabadi et al.12, 1978; Besserer et al.13, 1979; Williams and Zana14, 1980; Vogel and Yarborough15, 1980; Whitson and Torp16, 1983; Coats17, 1985; Coats and Smart18, 1986; Douglas and Bohle19, 1987; Drohm et al.20, 1988; Bette and Heinemann21, 1989; Chaback and Williams22, 1992; Furniual23, 1993). Some of these studies emphasize splitting the C7+ fraction as the key factor in attaining significant match between laboratory data and EOS results, Whitson24 (1983). The degree of splitting the C7+ fraction depends on the anticipated recovery process. Gas injection process with vaporizing gas phenomena require somewhat more splitting than depleted/water flooding processes. Others emphasize adjustment or regression of some of the EOS parameters to match the saturation pressure or the PVT data, Coats and Smart18 (1986) . Katz and Firoozabadi25 (1978) used methane interaction coefficient to study gas condensate behavior , dew point pressure and percentage liquid dropout. They correlated binaries between methane and petroleum fractions and used these binaries to predict dew point pressure for gas condensate. Peneloux and Jain26 (1979) used the EOS to predict the properties of gas condensate using three possibilities: 1) methane interaction only, 2) methane, ethane, and propane , 3) all the alkanes lighter than C6 as adjustable parameters of the EOS. They noticed that the observed difference between the results of using the three possibilities were not significant. Pederson et al.27 (1985) reported that setting the binary interaction (Ki,j) equal to zero for all nonhydrocarbon hydrocarbon components save computer time since usually their concentrations are less than 10 %. Coats and Smart18 (1986) studied the phase behavior of gascondensate using the EOS. They reported that EOS is generally not a good predictive tool for simulating the behavior of gas condensate and extensive splitting the C7+ fraction to match the laboratory data is generally unnecessary. Their results showed the poor predictive ability of the EOS without adjustment or regression of the equation of state parameters. These results

indicated that EOS capability improved when EOS parameters were regressed than when it were adjusted or the plus fraction were split. Whitson28 (1984) studied the effect of changing the critical properties and splitting the plus fraction on the EOS simulation of gas condensate. He showed that relatively small differences in critical properties and accentric factor can result in significant differences in EOS predictions. It is well known that a typical problem with using the EOS is the difficulty in describing petroleum fractions constituting C7+. These fractions significantly affects EOS predictions. Results of Whitson's work suggested that C7+ characterization has significant influence on EOS predictions of the reservoir fluid behavior. However, Whitson found that it is difficult to make objective conclusions about which C7+ characterization scheme is the best. He reported the temperature and compositional dependence of the binary interaction. Whitson found that saturation pressures may be overestimated if C1C7+ interaction matched at one temperature is used to estimate the saturation pressure at lower temperature. Recent years have witnessed a steady increase in the applications of artifical neural networks in petroleum Engineering, Ali29 (1994). Some examples of the application of neural networks are permeability prediction (Osborne30; 1992, Mohaghegh et al.31; 1996), identification of lithology (Zhou et al32 1993), drill bit diagnosis (Arehart33, 1989), and improvement in gas well production (Epping and Nitters34 1990). Neural network techniques have also been used in studying reservoir fluid phase behavior (Briones et al35, 1994; Habiballah et al36, 1996; Gharbi and Elsharkawy37,38, 1996,1997), interpretation of well testing data (Al-kaabi and Lee39, 1990; Juniard and Ershaghi40, 1993; Kumoluyi et al41, 1994; Sung et al42 , 1996), and determination of pore pressure (Accarian and Desbrandes43 , 1993). The objectives of this study can be summarized as follows: 1) Assess the accuracy of the EOS to determine the dew point pressure of gas condensate, in case it is not measured, using two different scheme of characterizing the plus fraction. 2) Simulate constant volume depletion performance of gas condensate using the adjusted or regressed EOS. Adjustment of the molecular weight of the plus fraction or regression of the EOS parameter is used to match the saturation pressure of the gas condensate as a first step for simulating the CVD. 3) Develop a neural network model capable of predicting the dew point pressure and simulating the depletion performance of gas condensate reservoirs using constant volume depletion data. 4) Compare the performance of the proposed network model and the simulated behavior of the equation of state to the experimentally measured data on constant volume depletion.

SPE 38855

EOS SIMULATION AND GRNN MODELING OF THE BEHAVIOR OF RETROGRADE-GAS CONDENSATE RESERVOIRS

Present study Constant volume depletion data of 142 gas condensate samples were available for this study. The data include two- phase gas deviation factor (Ztp), cumulative produced gas (Gp), and analysis of the produced gas condensate for a total of 1082 stages of depletion. Description of the data is given in Table (1). The data represents analysis of a world wide range of gas condensate systems collected world wide. CVD Constant volume depletion (CVD) is performed on gas condensates and volatile oils to simulate reservoir depletion process and compositional variation. Measured data can be used in material balance calculations and recovery process. Occasionally, when a gas condensate is first discovered, a reservoir fluid study is not always conducted. Thus, the compositional data of the CVD may not be available for reservoir simulation or material balance calculations. Therefore, EOS and correlations , if they are available, are used to predict fluid properties and composition at depleted stages. Generally, constant volume depletion test begins at dew point pressure (Pd). Volume at saturation pressure is used as a reference volume. In this test, pressure is reduced in steps, usually 4 to 10 steps, at reservoir temperature. After each pressure reduction, mercury is injected into the PVT cell and gas is withdrawn at constant pressure until the original volume is reached. Volume of gas withdrawn from the cell (Gp), compositional analysis of the produced gas, and liquid volume condensed in the cell (SL) expressed as a percentage of the reference volume are measured at each pressure, Niemtschik et al44 (1994). Volume of the produced gas at each pressure provides information about gas condensate recovery and condensed liquid volume represents the liquid saturation during depletion of gas condensate. The test also yields compressibility factor of the two phase (Ztp) present at each pressure depletion. Although laboratory experiment (CVD) form the cornerstone for determining fluid properties and depletion performance of gas condensate reservoir, such experiments can not be performed for all conditions. Therefore, EOS models and artificial neural network models, if they are available, can be used to predict performance of gas condensate reservoirs. EOS Model Soav-Redlich Kowng equation of state (SRK-EOS) is used in this study to calculate the dew point pressure and simulate the gas recovery and compositional changes during pressure depletion of gas condensate reservoirs. Calculation of dew point pressure Dew point pressures of 142 samples of gas condensate were calculated using SRK-EOS and compared with the experimentally measured dew point pressure using two methods of characterization of the plus fraction. For all the samples presented in this study, the C7+ fraction was divided

into C7, C8, C9, and C10+ using the splitting scheme described by Pedersen45 (1982). Two methods are considered in this study for characterizing the C10+ fraction of the gas condensate. The first method uses Cavett46 correlation (1964) to calculate critical pressure and critical temperature, Thomassen et al47 correlation (1986) and Kesler-Lee48 correlation (1976) to calculate the acentric factor for the plus fraction. In this study, this method is refered to as standard characterization. The second method uses correlations of Pedersen et al49 (1989) to calculate the critical properties and acentric factor. This method is refered to as the optional characterization. Simulation of CVD The first step in establishing the EOS parameters for studying depletion performance of gas condensate reservoir is to match the dew point pressure predicted by the equation of state to the experimentally measured one. This match could be achieved by adjusting the equation of state parameters, Characterization of the C7+ fraction (critical properties), adjustment of the molecular weight of the plus fraction, or controlling the binary interaction parameters (Kij). The binary interaction parameters are equal to zero for hydrocarbon-hydrocarbon interaction and non zero for interaction between a hydrocarbon and nonhydrocarbon and unlike pairs of non-hydrocarbons. Kij can be seen as indicator of discrepancy in size and polarity between two molecules. Table (2) shows the binary interaction coefficient used in this study. Two method are considered in this work for tunning the EOS parameters. The first method involves adjustment of the molecular weight of the plus fraction to match the measured dew point pressure. The second method emphasize regression of the EOS parameters to match experimentally measured dew point pressure. The adjusted or the regressed EOS at dew point is used latter to simulate compositional changes and recovery of gas condensate during pressure depletion for numerous samples from the data bank. Some of these samples are presented and discussed later in this study. General Regression Neural Network Models The present study introduces two general regression neural network (GRNN) models. The first model, GRNNM1, is used to estimate the dew point pressure and gas compressibility factor at the dew point pressure. The second model, GRNNM2, is used to estimate the composition and the cumulative production of the produced gas at any stage of pressure depletion for gas condensate reservoirs. GRNNM2 can also be used to estimate the initial composition of the gas condensate in case it is unavailable. GRNN General regression neural networks are universal single pass neural implementation of Parzens window based estimaters (Patzen50 1962) with one parameter, sometimes two, to adjust which makers training relatively, fast. The theory of such

ADEL M. ELSHARKAWY AND SALAH G. FODA

SPE 38855

networks can be developed as follows. Let x be a vector of random variables and y be a scalar random variable. Let X be a particular measured value of x. In other words, X is the finite vector of noisy measurements of x and y represents the associated scalar value. The conditional mean of y given X is given by

E[ y| X ] =

y f ( X , y ) dy / f ( X , y ) dy

(1)

where f(X,y) denotes the joint continuous probability density function (pdf) of a vector random variable X and a scalar random variable y. If f(X,y) is not known, an estimate f ( X , y ) must be used. Parzen50 (1962) proposed a class of consistent estimators for f(X,y). Using window estimation,

f ( X , y ) can be written as
f ( X , y) = 1 n(2 )
p +1

p +1

exp(( X X
i =1

i T

) ( X X i ) / 2 2 )
(2)

layers of parallel neural network architecture, Specht52,53 (1990,1991) where it called General Regression Neural Network. The GRNN is a three layer neural network with one hidden neuron for each training pattern. The number of neurons in the input layer (first slab) is the number of problem inputs and the number of neurons in the output layer (third slab) matches the number of outputs. The number of neurons in the hidden layer is usually the number of training patterns. The choice of higher number of hidden neurons may be advantageous in some problems. The GRNN training is achieved by comparing the input pattern in the n dimensional space of all training patterns to determine how far it is from these patterns. The predicted output is proportional to the distance of the given pattern from all training patterns. The distance metric is the familiar Euclidean distance. Also, the L1 - norms or the city block distance may be used. Even theory both norms are equivalent, it was found that city block distance may speed training. In conclusion, GRNN is a memory-bound three layer neural network which provides estimates of its variables and converges to the underlying linear or nonlinear regression surface. GRNN is more advantageous with sparse and noisy data than backpropagation (Specht and Shapiro54 1991 & Marquez55 1993) and is much faster to train. Using substantial simulations, Marquez (93) has shown that the GRNN sees through noise and distortion better than the backpropagation neural network. Compared with back propagates neural networks (BPNN), the BPNN performance can be significantly hindered by the presence of local minima and longer training. GRNNM1 Analysis of the initial composition of 117 samples (training samples) from the data bank described in this study is used to develop GRNNM1 to estimate the dew point pressure and gas compressibility factor at the dew point. Data of the remaining samples from the data bank (testing samples) are used to test the validity and accuracy of the proposed model. Input data to the GRNNM1 includes initial compositional analysis of the non-hydrocarbons (H2S, CO2, N2), hydrocarbons (C1 through C7+), molecular weight and specific gravity of the heptane plus fraction, and reservoir temperature. Outputs from the model are dew point pressure (Pd) and gas compressibility factor at the dew point (Zd). The designed GRNNM1, Figure (1), has three slabs, 15 neurons in the input slab, 180 neurons in the middle slab (about 33% more neurons), and two 2 neuron in the output slab with initial smoothing factor () of 0.3. The city block norm is used for convergence. GRNNM2 The GRNNM2 is designed to estimate the composition and cumulative produced gas at any stage of pressure depletion of gas condensate reservoirs. The model is developed using compositional analysis of 922 stages of pressure depletion of

exp (- (y - y i )2 / 2 2 )

p where x IR and n is the number of sample observations. Using (2), Eq. (1) becomes

y=

i =1 n i =1

e ( X X

i T

) ( X X i )/ 2 2

e
2 i

( X X i )T ( X X i )/ 2 2

ye

( y y i ) 2 / 2 2

dy

e
i

( y y i )2 / 2 2

dy
(3)

D = ( X X ) ( X X ) , then the estimate of the expected value y becomes


Let
i T

y( x ) = y e
i =1

i Di2 / 2 2

/ e Di / 2
2

(4)

i =1

These density estimators are consistent, i.e. they asymptotically converge to the underlying pdf f(x,y) at all points (x,y) provided that = (n) is chosen such that

lim ( n ) = 0 and lim n


n n

( n ) = . Therefore,

can be visualized as a smoothing parameter. That is, if is made large, the estimated density is forced to be smooth and 2 the limit becomes a multivariable Gaussian with covariance I. On the other hand, a smaller value of allows the estimated density to assume non-Gaussian shapes. In such case, wild points may have too great an effect on the estimate. Thus, given the joint probability density function (pdf) of X and y, the conditional pdf can be empirically determined and, hence, the expected value can be computed. Cacoullos51 (1966) extended Parzens method to the multivariable case. The resulting regression procedure is implemented via four

SPE 38855

EOS SIMULATION AND GRNN MODELING OF THE BEHAVIOR OF RETROGRADE-GAS CONDENSATE RESERVOIRS

gas condensate samples as a training set from the data bank. Compositional data of the rest of the depletion stages (testing set) are used for testing the validity, accuracy, and performance of the proposed model. The GRNNM2 uses dew point pressure, compressibility at dew point, reservoir temperature and reservoir pressure (or pressures) as input data to estimate the composition of the produced gas, nonhydrocarbons ( H2S, CO2, N2), hydrocarbons (C1 through C7+), molecular weight and specific gravity of the heptane plus, cumulative produced gas, and two phase compressibility factor for the remaining gas (Ztp). Thus, the model can be used to forecast constant volume depletion performance of gas condensate reservoirs in case expensive and laborious experimental data is unavailable. The proposed GRNNM2, Figure (2), has 3 slabs, 4 neurons in the input slab, 1500 neurons in the middle slab (about 60% more neurons), and 16 neurons in the output slab with initial smoothing factor of 0.3. The city block norm is also used for convergence. Later in this study, performance of the GRNNM2 in predicting the change in composition and cumulative produced gas for some of these samples is compared to EOS predictions. Results and discussion In this section calculations of dew point pressures and gas compressibility factors at dew point by EOS as well as by GRNNM1 are discussed. Also discussed in this section is the simulation of CVD using the EOS and modeling CVD behavior of gas condensate reservoir by GRNNM2. Dew point calculations Table (3) shows the accuracy of the dew point pressures calculated by EOS for the two methods of characterizing the plus fraction and the dew point pressures estimated by GRNNM1 for the training and testing samples. This table reports that the maximum deviation in calculated pressure by EOS is 1525 psi and maximum absolute error is 53% for the standard characterization, 1753 psi and 41% for the optional characterization, 93 psi and 1.8% for the training samples, and 723 psi and 37% for the testing samples. These results and others shown in Table (3) proves that GRNNM1 is successfully developed and trained to predict dew point pressure for gas condensate reservoir with better accuracy than using the EOS. Cross plots of measured dew point pressure versus calculated dew point pressure by EOS for the standard and optional characterization are shown in Figures (3) and (4), respectively, and crossplot of measured versus estimated dew point pressure by GRNNM1 is shown in Figure (5). Figures (3) and (4) show some of the plotted points around the unit slope line, however, Figure (5) shows almost perfect 45 degree line of all plotted points for the training and testing samples which indicates excellent agreement between measured pressures and estimated dew point pressures by GRNNM1. Table (4) reports errors in calculating gas compressibility factors at dew point pressures (Zdp) by EOS for the two

methods of characterizing the plus fraction as well as those by GRNNM1 for the training and testing samples. The maximum deviation in calculating (Zdp) by EOS is 0.364 for the standard characterization, 0.674 for the optional characterization, whereas the maximum deviation is 0.01 and 0.03 by GRNNM1 for the training and testing samples, respectively. Maximum deviation, Maximum absolute error percent (Emax), average relative error percent (Er), average absolute error percent (Ea), and correlation coefficient percent (R) reported in Table (4) show that GRNNM1 is properly trained to estimate Zdp more accurately than the EOS for the large range of samples that was used to develop the model and successfully tested to estimate Zdp for the rest of the samples. Cross plots of measured versus calculated gas compressibility factor at dew point by EOS for the standard and optional characterization are shown in Figures (6) and (7), respectively, and cross plot of measured versus estimated gas compressibility factor at dew point by GRNNM1 is shown in Figure (8). These plots show little scatter around the unit slope line for the Zdp by EOS methods and no scatter in the case of Zdp estimated by GRNNM1. GRNNM2 modeling and EOS Simulation of CVD tests Performance of the GRNNM2 to predict the changes in composition of the produced gas, cumulative produced gas, and two phase gas compressibility factor of the remaining gas at various stages of pressure depletion is discussed in this section. Also, some selected samples from the data bank are presented to show simulation results of CVD by EOS and GRNNM2 modeling of CVD tests. GRNNM2 modeling of CVD tests Table (5) shows accuracy of GRNNM2 in predicting compositional analysis of the produced gas, cumulative produced gas, and two phase compressibility factor of the remaining gas at various stages of depletion for the training samples as well as testing samples. The model successfully predicts composition of the produced gas, H2S through C7+ , of the training samples with a correlation coefficient of at least 99% and Max. Abs. error of less than 0.04% for all components except the heptane plus fraction. Thus, the model is properly trained and capable of capturing the physical trend of changing the composition and properties of the plus fraction as well as cumulative produced gas during CVD. The GRNNM2 is also able to forecast composition of produced gas, cumulative produced gas, and compressibility of the remaining two phase for all the depletion stages of the testing samples with a correlation coefficient of at least 95% except composition of C6 and specific gravity of C7+ (94% and 75%, respectively) and a maximum absolute error of less than 0.3 % for all samples except the molecular weight of the C7+ fraction. Summation of the mole fractions of all component of produced gas predicted by GRNNM2 is checked for training and testing samples at all depletion stages and found to equal to one. Thus the proposed model is accurate, valid and

ADEL M. ELSHARKAWY AND SALAH G. FODA

SPE 38855

reliable. The model can be used to forecast CVD data needed for many reservoir calculations in case this data is unavailable. GRNNM2 would save computer time and cost involved in predicting gas condensate behavior using EOS. Once properly trained, the model can be used to cut the expenses of frequent the sampling and the laborious experimental CVD test for gas condensate reservoirs. Examples showing the performance of GRNNM2 and EOS in predicting CVD results are presented below for some selected gas condensate reservoirs. These examples were chosen to represent samples from reservoirs that has the maximum percentage of C7+ , C1, H2S, CO2, and N2. Gas # 39 This gas is a rich gas condensate that contains 12.68% C7+ at dew point. Experimental results of constant volume depletion for this gas is shown in Table (6). Figures (9a) through (9d) show comparison between EOS simulations by adjusting the molecular weight of the C7+ (EOS-adj), EOS simulations by regression of the EOS parameters (EOS-reg), and general regression network modeling (GRNNM2) in matching experimentally measured C7+ content of the produced gas, methane content of the produced gas, cumulative produced gas (Gp), and two phase compressibility factor (Ztp) of the reservoir gas. Two conclusions can be drawn from these figures. First, GRNNM2 performed better than the EOS in matching the experimental data. Second, simulation results of CVD obtained by adjusting the molecular weight of the heptane plus fraction at dew point showed better agreement with experimental data than that when the EOS parameter were regressed to match the dew point.

Gas # 58 Compositional analysis of gas # 58 during CVD is shown in Table (6). This gas contains 12.51 % nitrogen at dew point. Simulation results of CVD of this gas by EOS and GRNNM2 are shown in Figure (12a) through (12d). Figure (12a) shows that neither the EOS nor GRNNM2 acceptably matched the experimental data of measured nitrogen content of the produced gas. Figure (12b) shows that GRNNM2 closely matched the measured C7+ content of the produced gas better than the EOS. However, Figures (12c) and (12d) indicate that the EOS matched the experimental data better than GRNNM2 for the two phase compressibility and the cumulative produced gas. Gas # 124 This gas contains 63.52 % CO2 at dew point. Constant volume depletion data for this gas is shown in Table (6). Figure (13a) through (13d) shows comparison between experimentally measured data for CO2, heptane plus, two phase compressibility factor and cumulative produced gas by GRNNM2 and EOS-reg and EOS-adj. Figure (13a) indicate that the EOS-reg and EOS-adj closely matched the experimental data better than GRNNM2. However the GRNNM2 shows better agreement with the experimentally measured C7+ content of the produced gas than the EOS-adj and EOS-reg, Figure 13(b). Simulation results of the two phase compressibility factor of reservoir gas (Figure 13c) and cumulative produced gas (Figure 13d) show that EOS and GRNNM2 equally matched the experimental data. Summary Two schemes of characterizing the plus fraction have been used in this study to calculate the dew point pressure by equation of state for gas condensate reservoirs. The study indicates that the optional characterization has a little improvement over the standard characterization. Several examples of simulating CVD for gas condensate reservoirs using the molecular weight of the plus fraction as a matching parameter for the dew point pressure showed better agreement between EOS and experimental data than when the EOS parameters were regressed to match the dew point pressure. The study also presents two general network models. The first model, GRNNM1 uses initial composition of gas condensate to predict dew point pressure and gas compressibility factor at dew . The second model, GRNNM2 uses dew point pressure, gas compressibility at dew point, reservoir temperature, and reservoir pressure (s) to predict composition of produced gas, cumulative production of produced gas, and two phase compressibility factor of remaining gas during pressure depletion of gas condensate reservoir. The models were trained and tested using CVD data of wide range of gas condensate samples collected world wide. Accuracy of the proposed models were checked by comparing their predicted performance for various gases to

Gas # 1 Gas #1 is a lean gas condensate containing 94.2% methane at dew point. Constant volume depletion data for this gas is shown in Table (6). Figures (10a) through (10d) show that GRNNM2 closely matched the methane content, heptane plus, two phase compressibility factor, and cumulative produced gas during pressure depletion of this gas.

Gas # 57 This gas has 28.16% H2S at dew point pressure of 4190 psi and 250 F. CVD data for this gas is shown in Table (6). Simulation results of CVD for this gas by EOS as well as by GRNNM2 are shown in Figures (11a) through (11d). Figures (11a) and (11b) indicate that GRNNM2 closely matched the experimental changes in H2S and C7+ ,respectively, better than the EOS-adj. Nevertheless, GRNNM2 and EOS-adj equally predict two phase compressibility factor of the reservoir gas (Ztp) and the cumulative produced gas (Gp).

SPE 38855

EOS SIMULATION AND GRNN MODELING OF THE BEHAVIOR OF RETROGRADE-GAS CONDENSATE RESERVOIRS

experimentally measured CVD and simulation results by equation of state. The study shows that these models are properly trained, and did successfully capture the physical trend of changing gas condensate properties during pressure depletion. The accuracy and reliability of the proposed GRNN models are demonstrated. These GRNN models can be used to forecast CVD needed for reservoir and production engineering calculations such as material balance calculations, reservoir simulations, separator design, and vertical performance calculations. The models can also be used to check accuracy of CVD tests and reduce time and money involved in simulations by EOS and laborious and expensive sampling. Acknowledgments The authors would like to thank the Kuwait Foundation for the Advancement of Sciences (KFAS) for providing financial support for this study, research grant N0.96-09-04. References 1- Conrard, P. G. and Gravier J. F. 1980. Peng-Robinson Equation of State Checks the Validity of PVT Experiments Oil & Gas Journal April 21, 72-86. 2- Raghavan, R and Jones JR. 1996 Depletion Performance of Gas Condensate Reservoirs, JPT August, 725-731. 3- Redlich, O. and Kwong, J. N. S. 1949. On the thermodynamics of solution V. An equation of state. Fugacities of gaseous solution. Chem. Review , 44, 233244. 4- Soav, G. 1972. Equilibrium constants from modified Redlich-Kwong equation of state. Chem. Eng. Sci., Vol. 27, 1197-1203. 5- Peng, D. Y. and Robenson, D. B. 1976. A new twoconstant equation of state. Ind. Eng. chem. Fund., vol. 15, No. 1, 59-64. 6- Zudkevitch, D. and Joeffe, J. 1970. Correlation and prediction of vapor liquid equilibria with the RedlichKwong equation of state., AICHE J., May, 16, 496-498. 7- Martine, 1979. Cubic Equation of state Which? Ind. and Eng. Chem. Fund., May, 81-80. 8- Penelox, A. Freeze P. 1982. A consistent correction to Kwong-Soave Volumes. Fluid Phase Equilib., 8, 7-23. 9- Saker, R. Danesh, A. S. and Todd A. C. 1991. Phase Behavior Modeling of Gas Condensate Fluids Using an Equation of State. Paper SPE 22714 presented at the Ann. Tech. Conf. & Exhibit. Dallas, TX Oct. 6-9. 10-Drohm, J. K. and Goldthrope W. H. 1988. On the Quality of Data from Standard Gas Condensate PVT Experiments. SPE 17768 Gas. Tech. Symp. Dallas, TX, June 13-15. 11-Al-Mahros, F. M. and Tijoa G. H. 1987. Analysis and Phase Behavior of Khuff Gas Condensate System in Bahrain field SPE 15766. Presented at the MEOS, Bahrain, March 7-10. 12-Firoozbadi, A., Hekim Y. and Katz D. L., 1978. Reservoir Depletion Calculations for Gas Condensate Using

Extended Analysis in the Peng Robinson Equation of State. J. Cand. Chem. Eng. Vol. 56, Oct. , pp. 610-615. 13-Besserer, G. J. and Serra, W., 1979. An Efficient Phase Behavior Package for Use in Compositional Reservoir Simulation Studies, SPE 8288. Ann. Tech. Conf. & Exhib. Los Vegas, Nevada, Sept. 23-26. 14-Williams C. A., and Zana, E. N. 1980. Use of the PengRobinson equation of state to predict hydrocarbon phase behavior and miscibility for fluid displacement. Paper SPE 8817. Enhance Oil Recovery Sym. Tulsa, OK, April 20-23. 15-Vogel, J. L., and Yarborough, L. 1980. Effect of nitrogen on the phase behavior and properties of reservoir fluids. SPE paper 8815, Enhanced oil Recovery Symp., Tulsa, Ok, April 20-23. 16-Whitson, C. H. and Torp S. B. 1983. Evaluating Condensate Volume Depletion Data. JPT March, 610620. 17- Coats K. H. 1985. Simulation of Gas Condensate Reservoir Performance, JPT Oct. 1870-1886. (1982 SPE 10512). 18-Coats K. H., and Smart G. T. 1986. Application of a Regression-Based EOS PVT Program to Laboratory Data. (1982 SPE 11197) SPE Reservoir Eng. May, 277-299. 19-Douglas, E. K., and Bohle, G. A. 1987. Third SPE Comparative Solution Project: Gas Cycling of Retrograde Condensate Reservoirs. JPT August 981-997. 20-Drohm J. K., Goldthrope, W. H., and Trengove R. 1988. Enhancing the Evaluation of PVT Data. OSEA 88174, 7th Offshore South East Asia, Feb., 1988, 626-635. 21-Bette, S. and Heineman, R. F. 1989. Compositional Modeling of High Temperature Gas Condensate Reservoir with Water Vaporization. SPE 18422. SPE Symp. on Res. Simulation., Houston, TX Feb. 6-8. 22-Chabak, J. J. and Williams, M. L., 1992. P-X Behavior of Rich Gas Condensate in admixture with CO2 and (N2 + CO2). SPE 24132 presented at the EOR Tulsa, pp. 22-24. 23-Furniual, S. R. 1993. Successful Prediction of Condensate Well bore Behavior Using EOS General from Black Oil Data. SPE 26683. Presented at Offshore European Conf., Aberdeen, Sept. 7-10. 24-Whitson, C. H. 1983. Characterizing Hydrocarbon Plus Fraction JPT August, 683-694. 25-Katz, D. L., Firoozabadi A. 1978. Predicting phase behavior of condensate/crude oil-systems using methane interaction coefficients. JPT, November, 1649-1655. 26-Penelox C., Jain E., 1979. Application of the developed Redlich-Kwong Equation of State to Predict the Thermodynamic Properties of Condensate Gases, SPE 8287. SPE Ann. Fall. Meet. Los Vegas, 27-Pederson K. S., Thomason P., and Fredenslund A. 1985. Thermodynamics of Petroleum. Mixtures Containing Heavy Hydrocarbons. Ind. Eng. Chem. Des. Dev. Vol. 24, No. 4, 948-954.

ADEL M. ELSHARKAWY AND SALAH G. FODA

SPE 38855

+ 28-Whitson, C. H. 1984. Effect of C7 Properties on Equation of State Prediction SPEJ, December, 685-696. 29-Ali, J. K. 1994. Neural networks: A new tool for the petroleum industry?. paper SPE 27561, European Per Conf., Aberdeen, 15-18 March. 30-Osborne, O. A. 1992. Neural networks provide more accurate reservoir permeability. Oil & gas J., Sept. 28, 8083. 31-Mohaghegh, et al. 1996. Petroleum reservoir characterization with the aid of artificial neural networks. J. Pet. Sci. Eng., Vol. 16, 26-274. 32-Zhou et al. 1993 determining reservoir properties in reservoir studies using a fuzzy neural network. paper SPE 26430, 68th Ann. Tech. Meet., Houston, TX, Oct. 3-6. 33-Arehart, R. A. , 1989. Drill bit diagnosis using neural networks. 1989 Conf. on Artificial Intelligence in Exploration and Production, Texas A&M. 34-Epping, W. J. and Nitters, G. 1990 Neural network for analysis and improvement of gas well production. Computer Simulation Conf., Calgary, July 16-18, 791796. 35-Briones, et al.. 1994. Application of neural network in the prediction of reservoir hydrocarbon mixture composition from production data. Paper SPE 28598, 69th Ann. Tech. Meet., New Orleans, Sept. 25-28. 36-Habiballah, et al. 1996. Use of neural networks for the prediction of vapor/liquid equilibrium K Values for light hydrocarbon mixtures. SPE Res. Eng., May, 121-125. 37-Gharbi, R. B. and Elsharkawy, A. M. 1996. Neural network model for estimating the PVT properties of Middle East crude oil systems. In Situ, Vol. 20, No. 4, 367-394. 38-Gharbi, R. B. and Elsharkawy, A. M. 1997. Universal neural network based model for estimating the PVT properties of crude oil systems. paper SPE 38099, Asia Pacific Conf., Kuala Lumpur, Malaysia, April 14-16. 39-Al-kaabi, A. W. and Lee, J. W. 1990. An artificial neural network approach to identify the well test interpretation model: application. paper SPE 2055, 65th Ann. tech. meet., New Orleans, LA, Sept. 23-26. 40-Juniardi, I. R. and Ershaghi, I. 1993. Complexities of using neural network in well test analysis of faulted reservoir. Paper SPE 26106, West Reg. Meet., Alaska, May 26-28. 41-Kumoluyi et al. 1994. Identification of well test models using high order neural net. paper SPE 27558, European Comp. Conf., Aberdeen, 15-17 March. 42-Sung et al. 1996 development of HT-BP neural network system for the identification of well test interpretation model SPE Comp. Appl., August, 102-105. 43-Accarian, P. and Desbrandes, R. 1993. Nuro-computing help pore pressure determination. Pet. Eng. Int. Feb. 3942. 44-Niemtschik, G. E., Poettmann, F. H., and Thompson R. S. 1994. New Correlation Determines Retrograde GasCondensate Composition. Oil and Gas J. Oct. 31, 33-38.

45-Pederson K. S., Thomason P., and Fredenslund A. 1982. Phase equilibria and separation process. Report SEP 8207, Inst. Kemiteknik, Denmark Terniske Hojskole. 46- Cavett, R. H. 1964. Physical data for distillation calculations, Vapor-liquid equilibria. 27th Midyear Meeting, API Division of refining. San Francisco, Ca, May 15, pp. 351-366. 47- Thomassen , P., Pedersen, K. S. and Fredenslund, A. A. 1986. Adjustment of C7+-molecular weights in the characterization of petroleum mixtures containing heavy hydrocarbons. SEP Report 8617, Institute for Kemiteknik, the Technical University of Denmark. 48- Kesler, M. G. and Lee, B. I. 1976. Improved prediction of enthalpy of fractions., Hydrocarbon Processing ,55, 153. 49- Pedersen, K. S., Thomassen, P. and Fredenslund, A. A. 1989. Advances in thermodynamics. 1, 137. 50- Parzen, E. 1962. On Estimation of a Probability Density Function and Mode, Annals of Mathematical Statistics, Vol. 33, pp. 1065-1076. 51- Cacoullos, T. 1966. Estimation of a Multivariable Density, Ann. Inst. Statist. Math (Tokyo), Vol. 18, No. 2, pp. 179-189. 52- Specht, D. F. 1990. Probabilistic Neural Networks, IEEE Trans. Neural Networks, Vol. 3, No. 1, pp. 109-118, January. 53- Specht, D. F. ``A General Regression Neural Network'', IEEE Trans. Neural Networks, Vol. 2, No. 6, pp. 568576, November. 54-Specht, D. F. and Shapiro. P. D. 1991 Generalization Accuracy of Probabilistic Neural Networks Compared with Backpropagation Networks, Proc. International Joint Conf. on Neural Networks, Seattle, WA, pp. 887892, July 8-14. 55-Marquez, L. and Hill, T. 1993. Function Approximation Using Backpropagation and General Regression Neural th Networks, Proc. 26 Hawaii International Conf. on System Sciences, Wailea, HI, pp. 607-615, January 5.

SPE 38855

EOS SIMULATION AND GRNN MODELING OF THE BEHAVIOR OF RETROGRADE-GAS CONDENSATE RESERVOIRS

Table 1- Description of the constant volume depletion data used in this study No. of sample 142 No. of depletion stags 1082
TF Min max Ave 94 327 241 H2S 0.0000 0.3182 0.0092 CO2 0.0000 0.6716 0.0256 N2 0.000 0.1284 0.0179 C1 0.1937 0.9473 0.7628 C2 0.0194 0.2092 0.0775 C3 0.0061 0.1299 0.0390 iC4 0.0017 0.0638 0.0082 Nc4 0.0025 0.0502 0.0139 iC5 0.0012 0.0431 0.0055 nC5 0.0008 0.0209 0.0055 C6 0.0012 0.0592 0.0067 C7+ 0.0022 0.1719 0.0279 Mw C7+ 98 253 127 SG C7+ 0.58 0.85 .77 Ztp 0.415 1.775 0.900 Press psi 200 11830 2864 Pd psi 555 11830 5232 Zd 0.704 1.775 1.079 Gp% 0.000 0.9221 0.3481

Table 2- Binary interaction coefficient N2 0.00 0.00 0.00 0.02 0.06 0.08 0.08 0.08 0.08 0.08 0.08 0.08 CO2 0.00 0.00 0.12 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.12 H2S 0.00 .012 0.00 0.08 0.07 0.07 0.06 0.06 0.06 0.06 0.05 0.05

N2 CO2 H2S C1 C2 C3 i C4 n C4 i C5 n C5 C6 C7+

10

ADEL M. ELSHARKAWY AND SALAH G. FODA

SPE 38855

Table (3) Error in estimating the dew point pressure by EOS and GRNNM1 Max Deviation, psi EOS Standard characterization Optional Characterization ANN Training data Testing data 1525 1753 EMin % 0.065 0.228 EMax % 53.079 41.336 Er % -6.427 1.182 Ea % 10.409 7.858 R% 94.783 95.067

93 723

0.000 0.121

1.852 37.71

0.024 -3.964

0.271 6.932

99.999 98.512

Table (4) error in estimating Z at the dew point by EOS and GRNNM1 Max Deviation EOS Standard characterization Optional Characterization ANN Training data Testing data EMin % 0.103 0.098 0.000 0.064 EMax % 44.722 94.134 0.966 6.782 Er % -2.792 -0.335 -0.005 -3.600 Ea % 5.374 5.353 0.133 1.968 R%

0.364 0.674 0.01 0.03

91.43 85.04 99.99 98.88

Table (5) accuracy of the GRNNM2 in estimating CVD analysis Training samples Ea % H2 S CO2 N2 C1 C2 C3 i-C4 n-C4 i-C5 n-C5 C6 C7+
MWC7+

Testing samples R% 99.6 99.9 99.9 98.8 99.5 99.5 99.8 99.5 99.7 99.5 99.1 98.8 99.9 99.8 99.8 99.4 Ea % 0.001 0.002 0.001 0.012 0.002 0.002 0.000 0.001 0.000 0.000 0.001 0.004 3.919 0.005 0.030 0.049 EMax % 0.027 0.087 0.012 0.218 0.025 0.019 0.004 0.016 0.005 0.009 0.027 0.073 53.625 0.138 0.276 0.286 R% 98.4 99.9 99.9 96.5 99.1 98.0 97.9 95.5 95.9 93.5 63.9 89.4 90.9 75.4 92.5 97.4

EMax % 0.029 0.016 0.005 0.082 0.028 0.015 0.002 0.005 0.002 0.002 0.009 31.147 7.474 0.022 0.077 0.422

SGC7+ Ztp Gp

0.001 0.000 0.000 0.003 0.001 0.001 0.000 0.000 0.000 0.000 0.000 0.075 0.607 0.001 0.005 0.013

SPE 38855

EOS SIMULATION AND GRNN MODELING OF THE BEHAVIOR OF RETROGRADE-GAS CONDENSATE RESERVOIRS

11

Table (6) Experimental data on constant volume depletion


Gas 1 IND 1 2 3 4 5 6 7 281 282 283 284 285 286 287 439 440 441 442 443 444 445 446 447 448 449 450 451 452 926 927 928 929 930 931 932 933 934 935 936 937 938 939 T(F) 221 221 221 221 221 221 221 313 313 313 313 313 313 313 250 250 250 250 250 250 250 293 293 293 293 293 293 293 219 219 219 219 219 219 219 209 209 209 209 209 209 209 H2S 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.2816 0.2767 0.2722 0.2695 0.2732 0.2892 0.3182 0.0215 0.0214 0.0214 0.0213 0.0213 0.0215 0.0221 0.0034 0.0034 0.0033 0.0033 0.0033 0.0034 0.0036 0 0 0 0 0 0 0 CO2 0.0017 0.0017 0.0017 0.0017 0.0018 0.0019 0.0020 0.0231 0.0242 0.0248 0.0253 0.0258 0.0262 0.0266 0.0608 0.0644 0.0669 0.0685 0.0694 0.0699 0.0679 0.0365 0.0367 0.0368 0.0370 0.0370 0.0370 0.0370 0.6352 0.6395 0.6514 0.6579 0.6639 0.6706 0.6716 0.0033 0.0033 0.0034 0.0035 0.0035 0.0036 0.0038 N2 0.0150 0.0156 0.0162 0.0168 0.0169 0.0164 0.0157 0.0137 0.0155 0.0161 0.0166 0.0163 0.0155 0.0143 0.0383 0.0455 0.0476 0.0473 0.0461 0.0434 0.0394 0.1251 0.1269 0.1277 0.1281 0.1284 0.1274 0.1249 0.0386 0.0399 0.0410 0.0417 0.0421 0.0411 0.0388 0.0032 0.0033 0.0033 0.0033 0.0033 0.0033 0.0033 C1 0.7284 0.7535 0.7735 0.7856 0.7945 0.7919 0.7776 0.6583 0.7074 0.7380 0.7559 0.7583 0.7485 0.7292 0.4033 0.4382 0.4641 0.4807 0.4844 0.4688 0.4331 0.6864 0.6914 0.6957 0.6988 0.6995 0.6991 0.6970 0.1937 0.1988 0.2008 0.2070 0.2084 0.2037 0.1994 0.9420 0.9438 0.9451 0.9461 0.9468 0.9473 0.9467 C2 0.0847 0.0850 0.0853 0.0856 0.0865 0.0883 0.0929 0.0803 0.0817 0.0821 0.0839 0.0863 0.0905 0.0944 0.0448 0.0471 0.0481 0.0487 0.0493 0.0496 0.0494 0.0481 0.0481 0.0482 0.0483 0.0485 0.0488 0.0491 0.0303 0.0307 0.0308 0.0309 0.0313 0.0315 0.0318 0.0231 0.0230 0.0230 0.0231 0.0232 0.0233 0.0236 C3 0.0418 0.0405 0.0393 0.0391 0.0395 0.0418 0.0452 0.0417 0.0411 0.0404 0.0402 0.0415 0.0447 0.0495 0.0248 0.0243 0.0239 0.0237 0.0239 0.0252 0.0277 0.0191 0.0190 0.0190 0.0189 0.0190 0.0192 0.0196 0.0174 0.0172 0.0170 0.0169 0.0170 0.0175 0.0184 0.0082 0.0082 0.0082 0.0082 0.0082 0.0082 0.0083 IC4 0.0110 0.0104 0.0100 0.0096 0.0094 0.0100 0.0114 0.0078 0.0073 0.0070 0.0069 0.0073 0.0082 0.0091 0.0060 0.0055 0.0051 0.0049 0.0049 0.0055 0.0067 0.0047 0.0046 0.0045 0.0044 0.0044 0.0046 0.0048 0.0033 0.0032 0.0031 0.0030 0.0030 0.0032 0.0035 0.0023 0.0023 0.0023 0.0023 0.0023 0.0023 0.0023 NC4 0.0171 0.0159 0.0149 0.0145 0.0145 0.0154 0.0171 0.0184 0.0170 0.0162 0.0159 0.0167 0.0186 0.0208 0.0132 0.0120 0.0111 0.0106 0.0106 0.0114 0.0140 0.0086 0.0084 0.0082 0.0081 0.0082 0.0084 0.0088 0.0093 0.0088 0.0085 0.0082 0.0082 0.0088 0.0097 0.0025 0.0025 0.0025 0.0025 0.0025 0.0025 0.0026 IC5 0.0088 0.0080 0.0073 0.0068 0.0066 0.0068 0.0074 0.0075 0.0067 0.0062 0.0060 0.0062 0.0070 0.0080 0.0079 0.0068 0.0060 0.0055 0.0053 0.0058 0.0074 0.0039 0.0038 0.0036 0.0035 0.0036 0.0038 0.0041 0.0039 0.0036 0.0033 0.0031 0.0030 0.0033 0.0039 0.0012 0.0012 0.0012 0.0012 0.0012 0.0012 0.0012 NC5 0.0084 0.0079 0.0074 0.0065 0.0060 0.0059 0.0066 0.0108 0.0097 0.0089 0.0084 0.0086 0.0096 0.0107 0.0081 0.0069 0.0060 0.0054 0.0052 0.0057 0.0071 0.0042 0.0040 0.0039 0.0038 0.0038 0.0040 0.0045 0.0047 0.0042 0.0038 0.0036 0.0035 0.0038 0.0046 0.0008 0.0008 0.0008 0.0008 0.0008 0.0008 0.0009 C6 0.0124 0.0108 0.0093 0.0082 0.0074 0.0075 0.0090 0.0116 0.0110 0.0103 0.0086 0.0078 0.0082 0.0092 0.0121 0.0096 0.0078 0.0066 0.0060 0.0063 0.0077 0.0064 0.0061 0.0058 0.0056 0.0056 0.0059 0.0065 0.0051 0.0049 0.0046 0.0042 0.0036 0.0030 0.0034 0.0014 0.0013 0.0013 0.0013 0.0013 0.0012 0.0013 C7+ 0.0707 0.0507 0.0351 0.0256 0.0169 0.0141 0.0150 0.1268 0.0784 0.0500 0.0323 0.0252 0.0230 0.0282 0.0991 0.0630 0.0412 0.0286 0.0217 0.0192 0.0214 0.0355 0.0296 0.0252 0.0222 0.0207 0.0203 0.0216 0.0551 0.0458 0.0324 0.0202 0.0127 0.0101 0.0113 0.0120 0.0103 0.0089 0.0077 0.0069 0.0063 0.0060 MW7+ 152 143 134 126 118 114 117 191 154 139 128 120 115 113 165 121 116 112 109 107 107 139 131 123 116 111 109 109 170 153 139 128 118 110 106 143 133 126 120 116 114 114 SG7+ 0.810 0.801 0.792 0.784 0.776 0.772 0.775 0.831 0.804 0.789 0.778 0.770 0.765 0.763 0.818 0.778 0.773 0.768 0.764 0.762 0.762 0.785 0.778 0.770 0.763 0.757 0.755 0.755 0.811 0.797 0.783 0.773 0.763 0.755 0.751 0.787 0.777 0.769 0.763 0.760 0.758 0.758 Ztp 0.997 0.929 0.876 0.835 0.807 0.767 0.679 1.212 1.113 1.021 0.932 0.844 0.739 0.612 0.838 0.788 0.75 0.718 0.686 0.639 0.553 1.066 1.017 0.979 0.951 0.936 0.927 0.902 0.851 0.78 0.729 0.718 0.735 0.75 0.72 1.019 0.971 0.942 0.928 0.93 0.942 0.949 P, psi 4973 4300 3600 2900 2100 1300 700 6010 5100 4100 3000 2000 1200 700 4190 3600 3000 2400 1800 1200 700 5150 4400 3600 2800 2000 1300 700 4825 4100 3300 2600 1900 1200 700 4786 4000 3300 2600 1900 1300 700 pd, psi 4973 4973 4973 4973 4973 4973 4973 6010 6010 6010 6010 6010 6010 6010 4190 4190 4190 4190 4190 4190 4190 5150 5150 5150 5150 5150 5150 5150 4825 4825 4825 4825 4825 4825 4825 4786 4786 4786 4786 4786 4786 4786 Zd 0.997 0.997 0.997 0.997 0.997 0.997 0.997 1.212 1.212 1.212 1.212 1.212 1.212 1.212 0.838 0.838 0.838 0.838 0.838 0.838 0.838 1.066 1.066 1.066 1.066 1.066 1.066 1.066 0.851 0.851 0.851 0.851 0.851 0.851 0.851 1.019 1.019 1.019 1.019 1.019 1.019 1.019 Gp 0.0000 0.0530 0.1762 0.3045 0.4717 0.6585 0.7909 0.0000 0.0755 0.1890 0.3490 0.5195 0.6693 0.7648 0.0000 0.0857 0.1984 0.3297 0.47Z84 0.6208 0.7423 0.0000 0.1035 0.2372 0.3885 0.5554 0.7071 0.8363 0.0000 0.0726 0.2009 0.3600 0.5419 0.7154 0.8255 0.0000 0.0000 0.0000 0.0000 0.0000 0.7035 0.8405

39

57

58

124

125

12

ADEL M. ELSHARKAWY AND SALAH G. FODA

SPE 38855

Input layer (15 Neurons)


T H2s Co2 N2 C1 C2 C3 IC4 NC4 IC5 NC5 C6 C7+ Mwc7+

Hidden Layer (180 Neurons)

Output Layer (2 Neurons)

Pattern And Summation Layer

Pd Zd

c7+

Figure 1- Architecture of GRNNM1

Output Layer (2 Neurons)


H2s Co2 N2 C1 C2 C3 IC4 NC4 IC5 NC5 C6 C7+ MWc7+

Hidden Layer (2 Neurons)


Pd Zd P T

Hidden Layer (1500 Neurons)

Pattern And Summation Layer

c7+ Ztp Gp

Figure 2- Architecture of GRNNM2

SPE 38855

EOS SIMULATION AND GRNN MODELING OF THE BEHAVIOR OF RETROGRADE-GAS CONDENSATE RESERVOIRS

13

EOS Estimated dewpoint pressure, ps

EOS Z @ Dewpoint Pressure

12000 10000 8000 6000 4000 2000 0 0 2000 4000 6000 8000 10000 12000 Measured dew point pressure, psi Fig 3. Crossplot of dew pint pressure ( Standard Characterization)

2.0 1.8 1.6 1.4 1.2 1.0 0.8 0.6 0.6 0.8 1.0 1.2 1.4 1.6 Experimental Z @ Dew Point Pressure Fig 6. Crossplot of Z @ dew point (Standard Characterization) 1.8

EOS Estimated dewpoint pressure, ps

10000 8000 6000 4000 2000 0 0 2000 4000 6000 8000 10000 12000

EOS Z @ Dew Point Pressure

12000

1.8 1.6 1.4 1.2 1.0 0.8 0.6 0.6 0.8 1.0 1.2 1.4 1.6 Experimental Z @ Dew Point Pressure Fig. 7 crossplot of Z @ dew pint (Optional Characterization) 1.8

Measured dew point pressure, psi Fig. 4 Crossplot of dew point pressure (Optional Characterization)

test data Trn data Estimated dew point pressure, psi 14000 Estimated Z @dew point 12000 10000 8000 6000 4000 2000 0 0 2000 4000 6000 8000 10000 12000 0.6 0.6 Measured dew point pressure, psi Fig. 5 Cross plot of dew point pressure, GRNN model 1.8 1.6 1.4 1.2 1.0 0.8

test data Trn data

1.0 1.2 1.4 1.6 1.8 Measured Z @ dew point Fig. 8 Cross plot of Z @ dew point , GRNN model

0.8

14

ADEL M. ELSHARKAWY AND SALAH G. FODA

SPE 38855

0.14 0.12 Mole fraction C7+ 0.10 0.08 0.06 0.04 0.02 0.00 0

Mole fraction C1

Exp Eos-reg Eos-adj GRNNM2

0.78 0.76 0.74 0.72 0.70 0.68 0.66 0.64 0 2000 4000 Pressure, psi 6000 8000 Exp Eos-reg Eos-adj GRNNm2 Figure 9b

Figure 9a 2000 4000 Pre ssure , ps i 6000 8000

0.90 0.80 0.70 0.60 0.50 0.40 0.30 0.20 0.10 0.00 0 2000 4000 Pressure, psi

Produced gas

Exp Eos-reg EOS-adj GRNNM2 Two phase Z

1.4 1.2 1.0 0.8 0.6 0.4 0.2 Figure 9d Exp Eos-reg Eos-adj GRNNM2

Figure 9c

6000

8000

0.0 0 2000 4000 Pressure, psi 6000 8000

Figure 9- Simulation of CVD for gas #39


0.82 0.80 0.78 0.76 0.74 0.72 0 1000 2000 3000 4000 Pressure, psi 5000 6000 Exp GRNNM2 EOS-adj EOS-Reg 0.08 Mole fraction C7+ 0.06 0.04 0.02 Figure 10b 0.00 0 1000 2000 3000 4000 5000 6000 Pressure, psi Exp GRNNM2 EOS-Adj EOS-Reg

Mole fraction C1

Figure 10a

1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 0

Produced gas

Exp GRNNM2 EOS-Adj EOS-Reg Two phase Z

1.0

0.9

0.8 Exp GRNNM2 EOS-adj EOS-Reg 5000 6000

0.7 Figure 10d 0.6

Figure 10c 1000 2000 3000 Pressure, psi 4000 5000 6000 0 1000

2000 3000 4000 Pressure, psi

Figure 10- Simulation of CVD for gas #1

SPE 38855

EOS SIMULATION AND GRNN MODELING OF THE BEHAVIOR OF RETROGRADE-GAS CONDENSATE RESERVOIRS

15

0.36 0.34 Mole fraction H2S 0.32 0.30 0.28 0.26 0.24 0.22 0.20 0 1000 2000 3000 4000 5000 Figure 11a Exp GRNNM2 EOS-Reg

0.50

Mole fraction C1

0.45

0.40 Exp GRNNM2 EOS-Reg 0.35 0 1000 2000 3000 4000 5000 Figur e 11b

Pressure, psi
1.00 Cumulative produced gas 0.80 0.60 0.40 0.20 0.00 0 1000 2000 3000 4000 5000 Ex p GRNNM2 EOS- Reg Two phase Z 1.0 0.8 0.6 0.4 0.2

Pr e s s ur e , ps i

Fig u r e 11c

Fig u r e 11d 0.0 0 1000 2000 3000

Ex p GRNM2 EOS-Reg 4000 5000

Pr e s s u r e , p s i

Pr e s s u r e , p s i

Figure 11- Simulation of CVD for gas # 57


0.130
0.04 0.04

Mole frcation N2

Mole fraction C7+

0.03 0.03 0.02 0.02 0.01 0.01 Figur e 12b Exp GRNNM2 EOS-adj 4000 6000

0.125

Exp GRNNM2 EOS-adj 0.120 0 2000

Figure 12a

4000

6000

0.00 0 2000 Pr e s s ur e , ps i

Pressure, psi

0.90 0.80 Cumulative produced gas 0.70 0.60 0.50 0.40 0.30 0.20 0.10 0.00 0 2000 4000 6000 Pressure, psi Figure 12c Exp GRNNM2 EOS-adj
Two phase Z

1.1 EOS-adj GRNNM2 EOS-adj

1.1

1.0

1.0 Figure 12d 0.9 0 2000 4000 6000 Pre s s ure , ps i

Figure 12- Simulation of CVD for gas 58

16

ADEL M. ELSHARKAWY AND SALAH G. FODA

SPE 38855

0.70 0.68 0.66 0.64 0.62 0.60 0.58 0.56 0.54 0.52 0.50 0

0.07 0.06 Mole fraction C7+ % 0.05 0.04 0.03 0.02 0.01 0.00 0 2000 4000 6000 Figure 13b Exp EOS-Reg EOS-adj GRNNM2

Mole fraction CO2

Figure 13a

Exp EOS-Reg EOS-adj GRNNM2

2000

4000

6000

Pressure, psi

Pressure, psi

0.90 Cumulative produced gas 0.80 0.70 0.60 0.50 0.40 0.30 0.20 0.10 0.00 0 1000 2000 3000 4000 5000 6000 Exp EOS-Reg EOS-adj GRNNM2 Two phase Z Figure 13 c

0.90 0.85 0.80 0.75 0.70 0.65 0.60 0 2000 4000 6000 Figure 13d Exp EOS-Reg EOS-adj GRNNM2

Pressure, psi

Pressure, psi

Figure 13- Simulation of CVD for gas # 124

Vous aimerez peut-être aussi