Bayesian belief networks

1 downloads 0 Views 471KB Size Report
Jul 21, 2011 - first studies on EOR screening can be considered as the publication by Taber et al. (Taber & Martin, 1983), which uses the available data on ...

SPE 143282 Bayesian Network Analysis as a Tool for Efficient EOR Screening

M.M.Zerafat1, Sh. Ayatollahi ∗1, N. Mehranbod2, D. Barzegari2 1 EOR Research Center, School of Chemical and Petroleum Engineering, Shiraz University, Shiraz, Iran 2 School of Petroleum and Chemical Engineering, Shiraz University, Shiraz, Iran [email protected] Copyright 2011, Society of Petroleum Engineers This paper was prepared for presentation at the SPE Enhanced Oil Recovery Conference held in Kuala Lumpur, Malaysia, 19–21 July 2011.

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 have not been reviewed by the Society of Petroleum Engineers and are subject to correction by the author(s). The material does not necessarily reflect any position of the Society of Petroleum Engineers, its officers, or members. Electronic reproduction, distribution, or storage of any part of this paper 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 SPE copyright.

Abstract. The main tool for screening of EOR techniques is generally based on the criteria presented in a variety of tables and graphs given in the literature. These data are derived from the basic theory of multiphase fluid flow through porous media, reservoir simulation, laboratory experiments and existing field-scale experiences. The purpose of this study is to develop a procedure capable of combining the data extracted from different sources ranging from worldwide field experiences to the existing tables into a unified expert system. This expert system is based on Bayesian network analysis in order to sort the proper EOR techniques for further assessment by economical and environmental criteria. A data bank has been gathered from worldwide EOR/IOR techniques and analyzed using data mining procedure which is then combined with extracted data from previously published screening tables. Bayesian network quantitative learning technique was applied to different data combinations from the data bank to train the network which is to serve as the expert system. The produced expert system is also applied to the gathered data pertaining to 10 Iranian southwest reservoirs. The results show that, CO2 flooding can be the most promising among various EOR techniques, which is in agreement with a previous work. According to this study, considering reservoir characteristics, and excluding the economic limitations, CO2 flooding is considered as the most efficient EOR method for Iranian carbonate reservoirs under study. The results show that Bayesian Belief network analysis can be successful in the prediction of proper EOR technique by providing sufficient Data to train the network.

1. Introduction The problem of EOR Screening for the purpose of selection and implementation of proper EOR techniques for a specific reservoir with special oil and rock properties is addressed in several papers as a guide for petroleum engineers. One of the first studies on EOR screening can be considered as the publication by Taber et al. (Taber & Martin, 1983), which uses the available data on screening to summarize EOR screening criteria in a single table and also presents EOR spectrum by simple ∗

Corresponding Author, P.O. Box 71345-1719 Shiraz, Iran

[email protected]

2

SPE 143282

graphical techniques. Guerillot (Guerillot, 1983), utilized artificial intelligence for the selection of proper EOR techniques. This developed expert system was able to propose reasonable descriptions for the selection of each EOR technique, based on fuzzy logic. The first computer based program was proposed by Parkinson et.al (Parkinson et.al, 1994), resulted in the technical selection of a few feasible techniques in the first place for further economical investigations. Taber et al. (Taber et al., 1997), proposed some criteria for most EOR techniques based on reservoir Data through the investigation of EOR mechanisms. Alvarado et.al (Alvarado et.al, 2002), collected a series of data on EOR screening projects all around the world and made use of space reduction techniques to show the existing correlations in parameter variations. Nowadays, with the aid of computers, artificial intelligence (AI) has become an inseparable part of engineering predictions and EOR screening has also been imparted this gift. Although the application of Artificial Intelligence techniques for chemical engineering and oil industry problems dates back to a couple of decades, few steps have been taken for the introduction of Bayesian Belief Networks in these fields. Various AI techniques have been examined for this purpose a few of which can be summarized as follows: The design of artificial neural networks (ANN) by Genetic Algorithm (GA) in the estimation of reservoir permeability is one of the AI applications in oil industry (Morooka et al., 2001). The application of GA for a full description of reservoir through estimation of other parameters such as porosity and recovery rate can be counted as another application in this field (Romero et al., 2001). The application of neural networks in the exploration and production is also examined in oil industry. The reservoir petro and geo-physical Data can be fed to ANN for characterization of fracture types and the estimation of pressure distribution in the reservoir (Greffioz et al., 1993). ANN has also been used in the determination of petroleum physical properties through the correlation of these parameters to other known variables such as temperature, pressure and natural boiling point (Aminzadeh, 2005). A series of publications by Shahab Mohaghegh (Mohaghegh et al., 2001, 2005; Rolon & Mohaghegh, 2009) investigates the application of several AI techniques in oil and gas reservoir characterization and reservoir log generation. Bayesian Network analysis which is commonly used in various fields for risk assessment (Biedermann et al., 2005) is also applied to a few petroleum industry problems (Carter et al., 2006). This technique has been applied successfully to predict asphaltene precipitation by using Data sources measured in laboratory and also gathered from literature (Sayyad Amin et al., 2010).

2. Bayesian Network Analysis The theory of Bayesian Belief Networks is based on the theory of probability which constructs the fittest framework for assessment in fields accompanied with indeterminacy. A Bayesian network can behave like human beings when confronting uncertainty and also provides a mathematical foundation to predict the likelihood of a target occurrence in future trials, from given occurrences in prior trials. The mere way to quantify a situation with an uncertain outcome is through determining its probability according to Bayesian logic. This Network consists of a series of variables and the relationships among them, each of which can be represented as a node in the network and possesses several states with no domains in common. Every directional link between two nodes shows the direct influence of one (parent) on another (child). This influence is quantified through a conditional probability distribution function correlating the states of each node with the states of its parents. The combination of variables and directional links forms a directional acyclic network (Fig.1). There exists a probability distribution accompanied with every child node (conditional probability) which depends on the probability distribution related to its own parents.

SPE 143282

3

A

B

C

E

F

D

G

Fig. 1 Example of an acyclic Bayesian Network Our information of the system and verified data enables us to calculate the new probability distributions for the nodes, not characterized in the network which is known as belief updating. The Bayes rule is the basis for updating process (Korb & Nicholson, 2004). Fig.1 represents a Bayesian Network for a joint probability distribution P (a, b, c, d , e, f , g ) . Node A , the root node, have no parents with (marginal) probability distribution P (a ) , where the domain of a is the set of values that A takes on with nonzero probability (prior probability). A key advantage of Bayesian networks is their synthesized representation of probabilistic relationships. In fact, it is necessary to consider only the known independencies among the variables in a domain, rather than specifying a complete joint probability distribution. The independencies declared at modeling time are then used to infer beliefs for all variables in the network (De Cristo et al., 2003) The limitation of Bayesian method is its subjectivity, especially for the establishment of prior belief. A few possible applications of Bayesian analysis can be amounted as: proposed risk assessment of nuclear waste disposal is calculated by Bayesian network (Lee et al., 2005). Bayesian approach is also used for a flow-field modeling to determine the greatest model uncertainty at the model boundaries (Abbaspour et al., 2000). Parameters estimation characterizing the hydrodynamic behavior of aquifers is done by Bayesian method (Ferraresi et al., 1996). Probabilistic expert systems are developed to produce a geographic distribution for the most probable sources of salinization by Bayesian belief networks methods (Ghabayen et al., 2006).

3. Data Collection In general, the Data available in the literature on EOR screening criteria fall into three main categories: (1) Laboratory tests have been done to check the feasibility of specific EOR process for known reservoir with fluids and rocks available. Experimental data available from these tests comprise the main core of Data usually used for EOR screening. (2) The second category is comprised of the Data produced from the simulation of oil reservoirs under EOR processes mostly by using commercial reservoir simulation software. (3) However, the most reliable category of information would be the specifications of reservoirs under successful EOR projects, whose technical and economical capabilities are proved practically. This last category can be considered the most reliable adequate information especially in AI techniques as it takes into account the whole essential criteria. Several EOR screening criteria in the form of tables have been proposed in the literature considering rock and fluid properties of oil reservoirs. The first table of this type is introduced by Taber et.al (Taber et. al. 1988), which

4

SPE 143282

suggests a special EOR technique in every given parameter range. Also, Taber has modified his table (Taber et. al. 1997) by using new data series and considering economical criteria in the investigations. Some EOR processes such as miscible CO2 flooding have found special attention and special tables are produced for them by some researchers (Klins 1984, Carcoana 1982). In another work (Picha, 2007) represents an updated screening criteria table for four EOR methods, considering the importance of CO2 flooding techniques and also the enhanced heavy oil production in recent years. Field Data used in this study is mostly extracted from a series of worldwide EOR surveys on field-scale successful EOR projects around the world (Worldwide EOR Survey, Oil & Gas J., 1994-2006). The gathered data are processed with data mining procedures for the selection of the most proper series in order to yield the best results in training process. Thorough investigation of gathered data proves the indeterminacy and similarities existent in various experiments which makes it necessary the utilization of an AI technique capable of decision making accompanied with probability and giving spectral answers rather than one-to-one conformity; the abilities inherent in Bayesian Analysis.

3.1 Effect of Reservoir Parameters on EOR Efficiency EOR process refers to techniques used to increase the oil recovery efficiency from oil reservoirs after primary and secondary oil recovery by using injection fluids such as different types of gases at different pressures or temperatures and various solvents. The injection may range from a few percents of reservoir volume to several pore volumes. In the cases involving an expensive injection, a limited amount of the fluid should be consumed due to economical considerations. If a large amount of injection fluid is required, the options will be limited to water and low price gases of avail. Basically, there are three mechanisms that control the efficiency of EOR processes and increase oil recovery factor rather than pure gas or water injection processes. These mechanisms are solvent extraction through miscibility, reducing interfacial tension and mobility control by decreasing oil viscosity or increasing water viscosity (Taber et. al. 1997). There are several EOR techniques which employ one or a combination of these mechanisms to enhance oil recovery. These techniques can be classified as follows based on the mechanisms considered above: 1. Miscible Gas Flooding; a) Miscible alcohol flooding b) Miscible hydrocarbon flooding c) Miscible CO2 flooding 2. Chemical Flooding; a) Surfactant flooding b) Alkaline flooding c) ASP flooding (Various Configurations) d) Polymer flooding 3. Thermal Oil Recovery; a) Continuous steam injection b) Cyclic steam injection c) Hot water injection d) In-situ combustion e) SAGD f) VAPEX

SPE 143282

5

4. New Methods; a) Microbial Flooding There are several general parameters which influence the selection of a special EOR process for a certain reservoir with unique rock and fluid properties. Also, there are some parameters only relevant to a particular EOR technique. Generally, the whole influential parameters can be categorized into two major categories. The first category is the reservoir properties and the second properties related to the reservoir fluid. General properties of the reservoir can be counted as porosity, permeability, reservoir formation type, reservoir depth, etc. On the other hand, oil density and viscosity are two examples of fluid properties. In addition, there are some unique properties in the investigation of particular EOR techniques which are presented in the next sections. Here, some important properties and their effect on various oil recovery processes are described. 3.1.1 Porosity Reservoir porosity accompanied by reservoir volume will determine the amount of oil in place. In a homogenous reservoir with fairly connected pores, the permeability of the reservoir rock is increased by increasing the porosity. In some cases such as polymer flooding, the porosity should be more than a definite value in order to be suitable for polymer permeation through the pore network. Fig. 2 presents the variations of porosity in different reservoirs under successful EOR production around the world. These EOR techniques are considered to be thermal techniques (including steam flooding, hot water injection and in situ combustion), miscible flooding (CO2 and hydrocarbon flooding) and immiscible flooding (CO2 and Nitrogen injection), polymer injection and microbial flooding respectively from left to right. As it is shown in Fig. 2 the average porosity is 22 % and porosity ranges from 1 % and 60 %. It is obvious that thermal EOR techniques entail comparatively higher porosities.

Fig. 2 The porosities of reservoirs under successful EOR processes in different parts of the world

6

SPE 143282

3.1.2 Permeability Increasing the rock permeability which results in decreasing the injection pressure leads to the improvement of production efficiency. There is a determined range for permeability that polymer molecules are blocked and can't permeate through the reservoir. The selection of an appropriate molecular weight distribution for polymer would decrease such a limitation. On the other hand, in some cases selective plugging using especial materials such as polymers are used to increase the possibility of oil production from un-swept zones. The disparity of reservoir permeability is shown in Fig 3. Although it seems there are many oscillations in the permeability variation range, but the major part of data varies between 1 to 10000 mD with average permeability as 870 mD.

Fig. 3 Permeabilities of reservoirs under successful EOR processes in different parts of the world 3.1.3 Reservoir formation type Observed formation types in reservoirs are usually sandstone, limestone, dolomite, conglomerate, tripolite and unconsolidated sand (Worldwide EOR Survey, Oil & Gas J., 1994-2006). The wettability of rocks and their pore structures are very dependent on the formation type. As it was mentioned before, formation type may affect wettability or wettability alteration during each production stage, which governs the next successful EOR/IOR processes. Among successful completed EOR projects using steam and hot water as the injection fluid almost all reservoirs were sandstone or unconsolidated sand reservoirs. Among successful in-situ combustion projects, sandstone reservoirs are the majority. Also, some successful in-situ combustion projects were carried out on dolomite reservoirs. Many polymer injection projects were performed on sandstone successfully. In other cases, all cited reservoir formation types were observed. It is worth mentioning that a reservoir may be composed of different formation types arranged in different production layers. 3.1.4 Reservoir depth Reservoir depth is an important factor to obtain miscibility in miscible flooding techniques and heat loss calculation in thermal oil recovery methods. Generally speaking, reservoir thickness should be less in thermal schemes. This issue causes the heat induced by hot injected fluid to cover the entire reservoir volume. In miscible enhanced oil recovery methods, reservoir thickness should be so high as to supply the minimum miscibility pressure. However steam flooding process is

SPE 143282

7

practical on reservoirs with depths falling between 500 and 5000 ft. The upper limit (500 ft) is considered due to the required steam pressure. The lower limit (5000 ft) is established because of high amount of heat loss from injection wells at high depths. By getting closer to 5000 ft, heat loss increases intensely even by utilizing properly isolating measures. In addition, as reservoir depth increases, the required injection pressure is increased as well. In such a condition, the condensable fraction of steam decreases and it results in the reduction of sweep efficiency. These variations are displayed in Fig. 4. As is obvious from Fig. 4, reservoirs depths under successful EOR projects range from 200 ft and 18500 ft and the average depth is 4715 ft.

Fig. 4 The depths variations of different reservoirs all around the world 3.1.5 Oil viscosity The heavier the reservoir oil, the less mobility is expected from reservoir fluid. This can lead to employment of thermal oil recovery techniques in order to reduce the oil viscosity. Decreasing oil viscosity causes to increase the mobility ratio and higher oil production. However, polymer injection entails the oil viscosity to be about 200 cp to be efficient. The higher the oil viscosity, the more concentrated the polymer has to be. The oil viscosity of different reservoirs may be variable from very low values to 800000 cp, while the mean oil viscosity is about 1870 cp. Fig. 5 displays oil viscosity variations. 3.1.6 Specific gravity (API) Oil specific gravity can be a representative of oil heaviness Similar to viscosity. In the same way, the lower API values, the more advisable thermal EOR would be. It should be mentioned that the API of different reservoir oils may vary between 1 to 60 degrees with the API mean around 28 (Fig. 6).

8

SPE 143282

Fig. 5 The viscosities of reservoirs under successful EOR processes in different parts of the world

Fig. 6 The specific gravities of reservoirs under successful EOR processes in different parts of the world 3.1.7 Reservoir temperature The higher the temperature of the reservoir, the lighter the reservoir oil may be. So it can be concluded that thermal oil recovery techniques are more suitable for the reservoirs with low temperatures. Reservoir temperature can be a criterion for surfactant and polymer flooding as well. These chemicals are too sensitive to high temperatures and they may decompose or suffer from a reduction in efficiency. Normally, the stability temperature of polymers is about 200 oF. Fig. 7 shows how reservoirs temperature varies for different reservoirs. The mean temperature is about 130 oF and temperature variation range is 10 to 325 oF.

SPE 143282

9

Fig. 7 The temperatures of reservoirs under successful EOR processes in different parts of the world

Fig. 8 The saturations of reservoirs under successful EOR processes in different parts of the world 3.1.8 Oil saturation Upon the exploitation of oil recovery techniques on a reservoir, the remaining oil in place is a function of the pervious processes and their degree of success. The remaining oil in place is very important for any feasibility and economical study to use specific types of EOR process. The variation of primary oil saturations for different reservoirs under successful EOR projects are shown in Fig. 8. Primary oil saturation for these reservoirs is variable between very low amounts up to 98 percent with average saturation as 60 %.

10

SPE 143282

4. Results and Discussion Two different sets of data are implemented for network training. First the Data produced from the Table named after Taber (Taber et al., 1997) and the second Data series extracted from reported successful EOR processes around the world. Firstly, the designed network is trained by the set of Data produced from Taber Table which consists of 5280 lines of Data. Then the trained network is tested by the same set of Data, the results of which are arranged in Table 1: Table 1. Test of trained network by the same Data set (Taber Table) N2-miscible

HC-miscible

CO2-miscible

Polymer

Combustion

Steam

Actual

64

0

0

0

0

0

N2-miscible

64

156

36

0

0

0

HC-miscible

64

156

36

0

0

0

CO2-miscible

0

0

0

669

191

4

Polymer

36

105

3

669

1290

201

Combustion

0

0

0

308

355

873

Steam

Error rate = 41.52%

The numbers on the main diagonal represent the number of Data predicted correctly as they were checked by Taber screening criteria. The reported error rate is a criterion of the amount of data domain entanglement. The Data rather than the main diagonal represent the number of wrong perditions. Another network is then trained by using the Data collected from world oil reservoirs under successful EOR processes, which consist of 1098 Data lines. The results of testing this trained network by the same Data set are shown in Table 2, which lacks an accurate prediction in most cases: Table 2. Test of trained network by the same Data set (Field Data) Steam

Combustion

CO2-miscible

CO2-immiscible

HC-miscible

Polymer

Actual

78

95

48

24

60

49

Steam

5

6

14

5

3

3

Combustion

46

52

35

56

62

73

CO2-miscible

16

25

8

9

10

17

CO2-immiscible

24

22

15

18

13

27

HC-miscible

20

24

28

17

17

18

Polymer

Error rate = 84.74%

The next step is the trial removal of different parameters in order to both perform parameter sensitivity analysis and also analyze the effect of saturation which turns to be unvarying in different EOR processes as is obvious in Fig. 7. Besides the porosity is not present in Taber Table and thus can’t be involved in the final data set. The miscible Nitrogen flooding is also removed from the set because of the few numbers of Data which can result in weak predictions by the network. The resting results are shown in Table 3 and some graphical results are presented in Figs. 8 to 11.

SPE 143282

11

Table 3. Test of trained network by the same Data set (Field Data without saturation) Steam

Combustion

CO2-miscible

HC-miscible

Polymer

Actual

346

1

0

0

4

Steam

7

25

1

0

3

Combustion

0

1

302

5

0

CO2-miscible

1

0

6

109

0

HC-miscible

11

1

14

1

88

Polymer

Error rate = 6.149%

Porosity Porosity1 0 Porosity2 0 Porosity3 100 Porosity4 0 25.5 ± 2.6

Temperature 0 Temp1 0 Temp2 0 Temp3 100 Temp4 240 ± 49

EO R Steam 21.6 Combustion 21.6 CO2miscible 21.3 HCmiscible 17.9 Polymer 17.5

Permeability Perm1 0 Perm2 100 Perm3 0 Perm4 0 106 ± 55

Viscosity 0 Viscosity1 0 Viscosity2 0 Viscosity3 100 Viscosity4 40300 ± 23000 Depth Depth1 0 Depth2 0 Depth3 100 Depth4 0 5450 ± 520

Gravity1 Gravity2 Gravity3 Gravity4

Permeability 0 Perm1 0 Perm2 100 Perm3 0 Perm4 600 ± 230

Viscosity 0 Viscosity1 0 Viscosity2 100 Viscosity3 0 Viscosity4 291 ± 170 Depth 0 Depth1 0 Depth2 0 Depth3 100 Depth4 12400 ± 3500

Gravity1 Gravity2 Gravity3 Gravity4

Viscosity 0 Viscosity1 0 Viscosity2 0 Viscosity3 100 Viscosity4 40300 ± 23000 Depth 0 Depth1 100 Depth2 0 Depth3 0 Depth4 3180 ± 790

Temperature 0 Temp1 0 Temp2 0 Temp3 100 Temp4 240 ± 49

EO R Steam 0.64 Combustion 0.72 CO2miscible 97.2 HCmiscible 0.71 Polymer 0.70

Gravity 0 0 0 100 49 ± 6.4

Fig 10. A typical answer of the trained network with field data with exclusion of saturation and porosity parameters. Porosity [21-30]; Perm. [200-1000]; Depth [6350-18500]; Gravity [36-60]; Viscosity [2.5-580]; Temp. [154-325], which predicts miscible CO2 injection with 97% probability.

Temperature 0 Temp1 0 Temp2 0 Temp3 100 Temp4 240 ± 49

EO R 16.9 Steam 22.7 Combustion CO2miscible 19.3 23.1 HCmiscible 18.0 Polymer

Permeability 0 Perm1 0 Perm2 0 Perm3 100 Perm4 50500 ± 29000

Gravity 100 0 0 0 7±4

Fig 8. A typical answer of the trained network with field data with exclusion of saturation and porosity parameters. Porosity [21-30]; Perm. [11-200]; Depth [6350-4550]; Gravity [0-14]; Viscosity [580-80000]; Temp. [154-325], which predicts almost the same probability for all methods.

Porosity 0 Porosity1 0 Porosity2 100 Porosity3 0 Porosity4 25.5 ± 2.6

Porosity 0 Porosity1 0 Porosity2 0 Porosity3 100 Porosity4 65 ± 20

Gravity1 Gravity2 Gravity3 Gravity4

Gravity 0 0 100 0 34.5 ± 2

Fig 9. A typical answer of the trained network with field data with exclusion of saturation and porosity parameters. Porosity [30-100]; Perm. [1000-100000]; Depth [1800-4550]; Gravity [31-38]; Viscosity [580-80000]; Temp. [154-325], which predicts almost the same probability for all methods.

Porosity 0 Porosity1 0 Porosity2 100 Porosity3 0 Porosity4 25.5 ± 2.6

EO R Steam 0.76 Combustion 0.74 CO2miscible 0.73 HCmiscible 0.73 Polymer 97.0

Permeability 0 Perm1 0 Perm2 100 Perm3 0 Perm4 600 ± 230

Temperature 0 Temp1 0 Temp2 100 Temp3 0 Temp4 132 ± 13 Viscosity 0 Viscosity1 0 Viscosity2 100 Viscosity3 0 Viscosity4 291 ± 170

Depth 0 Depth1 0 Depth2 100 Depth3 0 Depth4 5450 ± 520

Gravity 0 Gravity1 100 Gravity2 0 Gravity3 0 Gravity4 22.5 ± 4.9

Fig 11. A typical answer of the trained network with field data with exclusion of saturation and porosity parameters. Porosity [21-30]; Perm. [200-1000]; Depth [6350-4550]; Gravity [14-31]; Viscosity [2.5-580]; Temp. [110-154], which predicts polymer injection with 97% probability.

12

SPE 143282

Then the Data are analyzed by the removal of both saturation and porosity the results of which are presented in Table 4: Table 4. Test of trained network by the same Data set (Field Data without saturation and porosity) Steam

Combustion

CO2-miscible

HC-miscible

Polymer

Actual

345

0

0

1

4

Steam

10

19

3

1

3

Combustion

1

1

281

21

5

CO2-miscible

0

0

4

111

2

HC-miscible

18

2

21

1

73

Polymer

Error rate = 10.57%

The slight enhancement of error in prediction proves that porosity can be an influential parameter although we are obliged to omit it in the final Data set. Finally, the network is trained and tested by a combination of field and Taber Data the results of which are presented in Table 5: Table 5. Test of trained network by the same Data set (Field and Taber Data without saturation and porosity) Steam

Combustion

CO2-miscible

HC-miscible

Polymer

Actual

1862

19

1

0

8

Steam

909

1362

15

25

29

Combustion

0

11

60

46

2

CO2-miscible

1

10

25

269

18

HC-miscible

222

334

17

57

70

Polymer

Error rate = 32.56%

The results extracted from the network in different domains are presented in Figs. 12 to 17.

Permeability 0 Perm1 100 Perm2 0 Perm3 0 Perm4 106 ± 55

Depth 0 Depth1 100 Depth2 0 Depth3 0 Depth4 3180 ± 790

Steam Combustion CO2miscible HCmiscible Polymer

EO R 40.1 0.61 37.0 11.1 11.1

Gravity 0 Gravity1 100 Gravity2 0 Gravity3 0 Gravity4 22.5 ± 4.9

Temperature 100 Temp1 0 Temp2 0 Temp3 0 Temp4 50 ± 29

Viscosity 0 Viscosity1 0 Viscosity2 100 Viscosity3 0 Viscosity4 291 ± 170

Fig 12. A typical answer of the trained network with a combination of field and Taber Data excluding saturation and porosity parameters

Permeability Perm1 0 Perm2 100 Perm3 0 Perm4 0 106 ± 55

Depth Depth1 0 Depth2 0 Depth3 100 Depth4 0 5450 ± 520

Steam Combustion CO2miscible HCmiscible Polymer

Gravity1 Gravity2 Gravity3 Gravity4

EO R 1.38 1.36 48.0 48.0 1.34

Gravity 0 0 100 0 34.5 ± 2

Temperature Temp1 100 Temp2 0 Temp3 0 Temp4 0 50 ± 29

Viscosity Viscosity1 100 Viscosity2 0 Viscosity3 0 Viscosity4 0 0.5 ± 0.29

Fig 13. A typical answer of the trained network with a combination of field and Taber Data excluding saturation and porosity parameters which predicts both hydrocarbon and CO2on miscible techniques with the same probability.

SPE 143282

13

Permeability 0 Perm1 100 Perm2 0 Perm3 0 Perm4 106 ± 55

Steam Combustion CO2miscible HCmiscible Polymer

Depth 0 Depth1 0 Depth2 100 Depth3 0 Depth4 5450 ± 520

Temperature 0 Temp1 0 Temp2 100 Temp3 0 Temp4 132 ± 13

EO R 0.79 32.8 21.7 0.79 43.9

Gravity 0 Gravity1 100 Gravity2 0 Gravity3 0 Gravity4 22.5 ± 4.9

Depth Depth1 0 Depth2 100 Depth3 0 Depth4 0 3180 ± 790

Steam Combustion CO2miscible HCmiscible Polymer

Temperature Temp1 0 Temp2 100 Temp3 0 Temp4 0 105 ± 2.9

EO R 53.9 26.6 1.01 1.03 17.5

Gravity Gravity1 0 Gravity2 100 Gravity3 0 Gravity4 0 22.5 ± 4.9

Steam Combustion CO2miscible HCmiscible Polymer

Depth 0 Depth1 100 Depth2 0 Depth3 0 Depth4 3180 ± 790

Viscosity 0 Viscosity1 0 Viscosity2 100 Viscosity3 0 Viscosity4 291 ± 170

Fig 14. A typical answer of the trained network with a combination of field and Taber Data excluding saturation and porosity parameters.

Permeability Perm1 0 Perm2 100 Perm3 0 Perm4 0 106 ± 55

Permeability 0 Perm1 100 Perm2 0 Perm3 0 Perm4 106 ± 55

Gravity 0 Gravity1 100 Gravity2 0 Gravity3 0 Gravity4 22.5 ± 4.9

Viscosity 0 Viscosity1 0 Viscosity2 100 Viscosity3 0 Viscosity4 291 ± 170

Fig 15. A typical answer of the trained network with a combination of field and Taber Data excluding saturation and porosity parameters

Permeability Perm1 0 Perm2 0 Perm3 100 Perm4 0 600 ± 230

Steam Combustion CO2miscible HCmiscible Polymer

Depth Depth1 0 Depth2 0 Depth3 0 Depth4 100 12400 ± 3500

Viscosity Viscosity1 0 Viscosity2 0 Viscosity3 100 Viscosity4 0 291 ± 170

Fig 16. A typical answer of the trained network with a combination of field and Taber Data excluding saturation and porosity parameters.

Temperature 0 Temp1 0 Temp2 100 Temp3 0 Temp4 132 ± 13

EO R 28.4 19.9 6.62 6.62 38.4

Gravity1 Gravity2 Gravity3 Gravity4

Temperature Temp1 0 Temp2 0 Temp3 0 Temp4 100 240 ± 49

EO R 1.04 10.4 17.8 69.7 1.05

Gravity 0 0 0 100 49 ± 6.4

Viscosity Viscosity1 100 Viscosity2 0 Viscosity3 0 Viscosity4 0 0.5 ± 0.29

Fig 17. A typical answer of the trained network with a combination of field and Taber Data excluding saturation and porosity parameters.

Table 6. Test of trained network by 10% of the data Removed off the training set Steam

Combustion

CO2-miscible

HC-miscible

Polymer

Actual

122

4

0

0

0

Steam

62

72

3

1

18

Combustion

0

2

4

2

0

CO2-miscible

0

1

2

18

0

HC-miscible

15

21

1

4

6

Polymer

Error rate = 37.99%

In order to check the network ability for correct prediction through untrained domains, the network is tested by 10% of the Data removed off the training Data Set. The results are presented in Table 6 which turns out to be acceptable. The produced expert system is also applied to the gathered data pertaining to 10 Iranian southwest reservoirs. The results show that, CO2 flooding can be the most promising among various EOR techniques, which is in agreement with a previous work (Kord et al., 2008). According to this study, considering reservoir characteristics, and excluding the economic limitations, CO2 flooding is considered as the most efficient EOR method for Iranian carbonate reservoirs under study.

5. Conclusions The number of data available turns out to be determining and lack of sufficient data results in wrong predictions, particularly through untrained domains. In order to check the network ability for correct prediction through untrained domains, the network is tested by 10% of the Data removed off the training Data Set. The results are acceptable relative to the number of Data used for training.

14

SPE 143282

The produced expert system is also applied to the gathered data pertaining to 10 south-west Iranian reservoirs. The results show that, CO2 flooding can be the most promising among various EOR techniques.

References Abbaspour, K., Matta, V., Huggenberger, P., Johnson, C.A., 2000. A contaminated site investigation: comparison of information gained from geophysical measurements and hydro-geological modeling, J. Contaminant Hydrology. 40, 365-380. Abolhoseini, H., 2004. Study on Ultimate Heavy Oil Recovery from Zaqeh Field in S.W of Iran, SPE 88861, 11th Abu Dhabi international Petroleum Exhib. & Conf., Abu Dhabi, U.A.E., 10th-13th Oct. Al-Bahar, M.A., Merrill, R., Peake, W., Jumaa, M., Oskui, R., 2004. Evaluation of IOR Potential within Kuwait, SPE 88716, 11th Abu Dhabi international Petroleum Exhib. & Conf., Abu Dhabi, U.A.E., 10th-13th Oct. Alkafeef, S.F., Alforgi, M.Z., 2007. Review of and Outlook for Enhanced Oil Recovery Techniques in Kuwait Oil Reservoirs, IPTC 11234, International Petroleum Technology Conf., Dubai, U.A.E., 4th-6th Dec. Alvarado V., Ranson A., Hernandez K., Manrique E., Matheus J., Liscano T., Prosperi N., 2002. Selection of EOR/IOR Opportunities Based on Machine Learning, SPE 78332, 13th European Petroleum Conf., Aberdeen, U.K., 29–31. Aminzadeh,F., 2005. Applications of AI and soft computing for challenging problems in the oil industry, J. Pet. Sci. & Eng., 47, 5-14. Biedermann, A., Taroni, F., Delemont, O., Semadeni, C., Davison, A.C., 2005. The evaluation of evidence in the forensic investigation of fire incidents. Part II. Practical examples of the use of Bayesian networks, Forensic Science International, 147, 59-69. Carcoana, A., 1982. Enhanced Oil Recovery in Romania, Proceedings of 3rd Joint SPE/DOE Symposium on Enhanced Oil Recovery, SPE of AIME, Dallas, TX, 367-379. Carter, J.N., Ballester, P.J., Tavassoli, Z., King, P.R., 2006. Our calibrated model has poor predictive value: An example from the petroleum industry, Reliability Eng. & Sys. Safety. 91, 1373-1381. De Cristo, M.A.P., Calado, P.P., Silva, M.L.S.I., Muntz, R., Ribeiro-Neto, B., 2003. Bayesian belief networks for IR, International J. Approximate Reasoning. 34, 163-179. Ferraresi, M., Todinia, E., Vignolib, R., 1996. A solution to the inverse problem in groundwater hydrology based on kalman filtering, J. of Hydrology. 175, 567-581. Ghabayen, S.M.S., McKee, M., Kemblowski, M., 2006. Ionic and isotopic ratios for identification of salinity sources and missing data in the gaza aquifer, J. of Hydrology. 318, 360–373. Greffioz J., Olver, A.J., Schirmer, P., 1993. TOPEX: an expert system for estimating and analyzing the operating costs of oil and gas production facilities, SPE 25315, Asia Pac. Oil Gas Conf., Singapore, Feb. 8th–10th Guerillot, D.R., 1983. EOR Screening With an Expert System, SPE 17791 Hunedi, S., Danquigny, J., Morel, D., Maubeuge, F., Guarrino, A. 2005. Applicability of Enhanced Oil Recovery Techniques on Mature Fields-Interest of Gas Injection, SPE 93368, 14th SPE Middle East Oil & Gas Show and Conf., Bahrain International Exhibition Centre, Bahrain, 12th-15th March

SPE 143282

15

Klins, M.A., 1984. Carbon Dioxide Flooding: Basic Mechanisms and Project Design, International Human Resources Development Corporation, Boston, Massachusetts, 267-275. Korb, K.B., Nicholson, A.E., 2004. Bayesian artificial intelligence, Chapman & Hall/CRC Kord, Sh., Zerafat, M.M., Ayatollahi, Sh., Zobeidi, K., 2008. Screening of Iranian Carbonate Petroleum Reservoirs for CO2 injection, 12th Iranian Chem. Eng. Congress (IChEC12), Tabriz, Iran, Sahand Uni. Tech., 21th-23th Oct. Lee, C.J., Lee, K.J., 2005. Application of bayesian network to the probabilistic risk assessment of nuclear waste disposal, Reliability Engineering & System Safety. 9, 1-18. Mathiassen, O. M., CO2 as Injection Gas for Enhanced Oil Recovery and Estimation of the Potential on the Norwegian Continental Shelf, NTNU – Norwegian University of Science and Technology Department of Pet. Eng. & App. Geophy. Mohaghegh, Sh.D., Richardson, M., Ameri, S., 2001. Use of intelligent systems in reservoir characterization via synthetic magnetic resonance logs, J. Pet. Sci. & Eng. 29, 189–204. Mohaghegh, Sh.D., 2005. A new methodology for the identification of best practices in the oil and gas industry using intelligent systems, J. Pet. Sci. & Eng. 49, 239-260. Morooka, C.K., Guilherme, I.R., Mendes, J.R.P., 2001. Development of intelligent systems for well drilling and petroleum production, (2001). 32, 191-199. Neapolitan, R.E., 1990. Probabilistic Reasoning in Expert Systems: Theory and Algorithms, John Wiley & Sons, New York. Parkinson, W.J., Luger, G.F., Bretz, R.E., Osowski, J., 1994. Using an expert system to explore enhanced oil recovery methods, com. elect. Eng. 20, 2, 181-197. Pearl, J., 1991. Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference, Morgan Kaufmann, San Mateo, CA. 2nd Ed. Picha, M.Sh., 2007. Enhanced Oil Recovery by Hot CO2 Flooding, SPE 10542, 15th SPE Middle East Oil & Gas Show and Conf., Bahrain International Exhibition Centre, Bahrain, 11th-14th March Rolon, L., Mohaghegh, Sh. D., Ameri, S., Gaskari, R., McDaniel, B., 2009. Using artificial neural networks to generate synthetic well logs, J. Natural Gas Sci. & Eng. 1, 118-133. Romero, C.E., Carter, J.N., 2001. Using genetic algorithms for reservoir characterization, J. Pet. Sci. & Eng. 31, 113–123 Sayyad Amin, J., Alamdari, A., Mehranbod, N., Ayatollahi, Sh., Nikooee, E., 2010. Prediction of Asphaltene Precipitation: Learning from Data at Different Conditions, Energy & Fuels. 24, 4046-4053. Spiegelhalter, D.J., Dawid, A.Ph., Lauritzen, S.L., Cowell, R.G., 1993. Bayesian analysis in expert systems, Stat. Sci. 8, 3, 219-283. Taber, J.J., Martin F.D., 1983. Technical Screening Guides for the Enhanced Recovery of Oil, SPE 12069, 58th Annual Technical Conf. & Exhib., San Francisco, CA, 5th-8th Oct. Taber, J.J., Martin, F.D., Seright, R.S., 1997. EOR Screening Criteria Revisited - Part 1: Introduction to Screening Criteria and Enhanced Recovery Field Projects, SPE 35385, SPE Reservoir Engineering Journal, 189-198.

16

SPE 143282

Enhanced Oil Recovery Potential in the United States, 1978. Office of Technology Assessment, Congress of the United States, NTIS Order # PB-276594; 12 Vols. 1996 worldwide EOR survey, 1996. Oil & Gas J. 94, 16, 45. 1998 worldwide EOR survey, 1998. Oil & Gas J. 96, 16, 59. 2002 worldwide EOR survey, 2002. Oil & Gas J. 100, 15, 71. 2004 worldwide EOR survey, 2004. Oil & Gas J. 102, 14, 53. 2006 worldwide EOR survey, 2006. Oil & Gas J. 104, 15, 45.

Suggest Documents