Total organic carbon content determined from well logs using ΔLogR and Neuro Fuzzy techniques

Total organic carbon content determined from well logs using ΔLogR and Neuro Fuzzy techniques

Journal of Petroleum Science and Engineering 45 (2004) 141 – 148 www.elsevier.com/locate/petrol Total organic carbon content determined from well log...

1MB Sizes 0 Downloads 22 Views

Journal of Petroleum Science and Engineering 45 (2004) 141 – 148 www.elsevier.com/locate/petrol

Total organic carbon content determined from well logs using DLogR and Neuro Fuzzy techniques Mohammad Reza Kamalia,*, Ahad Allah Mirshadyb b

a Research Institute of Petroleum Industry, Tehran, Iran National Iranian Oil Company, Exploration Directorate, Iran

Received 30 December 2003; accepted 25 August 2004

Abstract Total organic carbon (TOC) content present in potential source rocks significantly affects the response of several types of well logs. Wireline logs can be used to identify source rocks and serve as an indicator for the source rock potential. Because the source rock intervals generally show a lower density, higher sonic transit time, higher porosity and higher resistivity than other sedimentary layers, these phenomena can be used to identify source rocks on wireline logs. This paper attempts to establish a quantity correlation between standard well logs (sonic, density, neutron and resistivity) and total organic carbon by means of two techniques, namely, using DLogR and then using Neuro Fuzzy. Two pilot wells bAQ and bBQ, through Pabdeh and Gurpi formations in the Dezful Embayment, were selected for this study. D 2004 Published by Elsevier B.V. Keywords: Total organic carbon content; DLogR; Neuro Fuzzy

1. Introduction Source rocks are commonly shales and lime mudstones that contain significant amounts of organic matter (Tissot and Welte, 1984). The feasibility of interpreting organic matter from wireline measurements comes from its physical properties, which differ considerably from those of the mineral components of its host rock: lower density, slower sonic velocity or higher sonic transit time, frequently higher uranium * Corresponding author. Tel./fax.: +98 21 5699410. E-mail address: [email protected] (M.R. Kamali). 0920-4105/$ - see front matter D 2004 Published by Elsevier B.V. doi:10.1016/j.petrol.2004.08.005

content, higher resistivity and higher hydrogen and carbon concentrations. Consequently, the logs used for source rock evaluations most commonly include density, sonic, gamma ray, neutron and resistivity (Serra, 1986; Herron, 1988; Luffel, 1992). Numerous studies have illustrated the potential value of wireline logs for source rock evaluation. Beers (1945), Swanson (1960), Fertle (1988), Schmoker (1981) and Hertzog et al. (1989) used gamma-ray spectral log for identifying organic-rich rocks. Schmoker and Hester (1983) proposed the use of the density log for estimating organic matter content. Dellenbach et al. (1983) and Hussain (1987)

142

M.R. Kamali, A.A. Mirshady / Journal of Petroleum Science and Engineering 45 (2004) 141–148

developed a method using the transit-time and gamma-ray curves to provide a parameter that relates linearly to organic richness. A method involving a combination of resistivity, density and sonic logs has been introduced by Meyer and Nederlof (1984). This method discriminates between source rocks and nonsource rocks without attempting to quantify the organic richness from combination of logs. Mendelson and Toksoz (1985) applied multivariate analysis of log data to characterize source rocks. At last, Passey et al. (1990) invented a new technique called DLogR. This technique employs the overlaying of porosity logs (sonic, density and neutron) and resistivity log for identifying and calculating total organic carbon. This paper has generally focused on the quantification of organic matter using DLogR and Neuro Fuzzy techniques.

2. The source rock A petroleum source rock is defined as any rock that has the capability to generate and expel enough hydrocarbons to form an accumulation of oil or gas. The most important factor controlling the generation of oil and gas is the hydrogen content of the organic matter (OM) (Hunt and Jaieson, 1956; Hunt, 1996). The quantity of organic matter usually is expressed as total organic carbon (TOC) and is measured with Rock-Eval technique. This technique provides several measurements. The first peak (S1) represents milligrams of hydrocarbons that can be

thermally distilled from 1 g of the rock. The second peak (S2) represents milligrams of hydrocarbons generated by pyrolytic degradation of the kerogen in 1 g of rock. The third peak (S3) represents milligrams of carbon dioxide generated from a gram of rock, and the last peak, S4, represents quantity of CO2 produced by oxidation of the residual organic matter and TOC, which is sum of residual organic carbon and pyrolyzed organic carbon (Espitalie et al., 1985; Peters, 1986; Langford, 1990).

3. Neuro Fuzzy system Neuro Fuzzy systems integrate fuzzy logic and neural networks; the architecture of the Neuro Fuzzy classifier is slightly different from the architecture used in function approximators (Tommi, 1994). The two first layers have the identical function with the approximator. Fig. 1 shows a Neuro Fuzzy system using the following fuzzy rules, Rule 1: If x1 is A1 and x2 is B1, then class is 1. Rule 2: If x1 is A2 and x2 is B2, then class is 2. Rule 3: If x1 is A1 and x2 is B3, then class is 1. Layer 3. Combination of firing strengths: If several fuzzy rules have the same consequence class, this layer combines their firing strengths. Usually, the maximum connective (or operation) is used. Layer 4. Fuzzy outputs: In this layer, the fuzzy values of the classes are available. The values describe how well the input of the system matches to the classes.

Fig. 1. Neural architecture of the Neuro Fuzzy classifier.

M.R. Kamali, A.A. Mirshady / Journal of Petroleum Science and Engineering 45 (2004) 141–148

Layer 5. Defuzzification: If the crisp classification is needed, the best-matching class for the input is chosen as output class.

143

present in seawater together with other trace elements, and that uranium thus becomes concentrated in source rock. 4.2. Resistivity logs

4. Log responses in source rocks The following well logs were available for research. 4.1. Gamma-ray log Organic-rich rocks can be relatively highly radioactive, i.e. they can have a higher gamma-ray reading than nonsource shales and limestones. This natural radioactivity is usually due to uranium, thorium and potassium enrichment. It can be postulated that plankton absorbs uranium ions that are generally

Source rocks are generally laminated and thus are electrically anisotropic. This increases the resistivity measured by spherically focused logs. When source rock becomes mature, free oil is present in voids and fractures, and with maturity, the resistivity of source rock increases significantly (Morel, 1999; Tavish, 1998). This makes it possible to use resistivity as a maturity indicator for a given source rock formation. Fig. 2(A) shows the deep resistivity vs. TOC in the Pabdeh and Gurpi formations for the well bAQ indicated. In as much as TOC content is electrically nonconductive, high TOC content can increase the

Fig. 2. (A) Plot of resistivity vs. measured TOC; (B) plot of bulk density vs. measured TOC; (C) plot of sonic vs. measured TOC; (D) plot of neutron porosity vs. measured TOC.

144

M.R. Kamali, A.A. Mirshady / Journal of Petroleum Science and Engineering 45 (2004) 141–148

resistivity of the host above the resistivity value of the same rock devoid of TOC. 4.3. Formation density logs This log measures the bulk density of the formation. This density consists of the combined effects of matrix density and fluid density. In other words, the more fluid a formation contains, the more porous it is. In shales with a similar degree of compaction and similar matrix and fluid density, water saturation should be equal. Solid organic matter has a similar density to that of water (approx. 1.0 g/ ml) and thus less than that of the surrounding rock matrix. If the density read in source rocks is lower than the density read in normal shales, it must be a function of the amount of organic matter, which is present. Fig. 2(B) displays the density value vs. TOC value in the Pabdeh and Gurpi formations in well bA.Q It is apparent that density decreases when TOC amount increases. 4.4. Sonic log The interval transit time (Dt) is the reciprocal of the velocity of the compressional wave and a function of formation lithology, porosity and the type and distribution models of fluids (water, gas, oil, kerogen, etc.). TOC content tends to increase the apparent Dt value. Fig. 2(C) shows the relationship between TOC and acoustic log responses in the Pabdeh and Gurpi formations in well bA.Q As expected and illustrated in Fig. 2(C), TOC wt.% increases with higher Dt values. 4.5. Neutron log This log essentially measures hydrogen concentration. Neutron log porosity responses are higher in source rocks than in nonsource rocks. As shown in Fig. 2(D), a neutron log shows a direct relationship with TOC wt.%.

5. Determining TOC from well logs For this purpose, two techniques were used: DLogR and Neuro Fuzzy.

5.1. DLogR technique This technique includes three methods using sonic/ resistivity, neutron/resistivity and density/resistivity logs overlay for TOC definition. In order to apply this technique in wells bAQ and bB,Q the amount of level of organic metamorphism (LOM) needs to be defined. In accordance with experimental formula, TOC content was calculated. This content was compared with the data obtained from experimental analysis of samples, and the error value was estimated using statistical software. 5.1.1. Use of sonic and resistivity log Resistivity log, with a registered range 0.2–2000 V m in logarithmic scale, and sonic log, which ranges from 140 to 40 As/ft in linear scale, show a good overlay in well bA.Q This occurred when changing range of special resistivity log is considered from 0.01–1000 V m and the sonic log is 140– 15 As/ft. The algebraic expression that was used by Exxon for the calculation of DLogR from the sonic/ resistivity is: DLogR ¼ logð R=Rbaseline Þ  PðDt  Dtbaseline Þ where DLogR is the curve separation measured in logarithmic resistivity cycles, R is the resistivity measured in V m by the logging tool, Dt is the measured transit time in As/ft, R baseline is resistivity corresponding to the Dt baseline value when the curves are baseline in nonsource rocks and P is based on the ratio of transit time cycle amount per one resistivity cycle. In well bA,Q Dt baseline=65 As/ft, and R baseline=10 V m and P=0.02 were selected. Exxon used the following empirical equation for calculating TOC in source rocks from DLogR: TOC ¼ DLogRT10ð2:2970:1688LOM Þ According to calculated DLogR and LOM=11 definition in accordance with experimental results obtained from geochemical analysis of samples, the TOC amounts were calculated and shown in Fig. 3 (Hood et al., 1975). Calculation of TOC in this method compared to results of TOC from geochemical analysis of samples is shown in Fig. 4. So, the mean square error obtained is 0.098. Similar to the method used for the sonic/resistivity overlay, the mean square error obtained for neutron/

M.R. Kamali, A.A. Mirshady / Journal of Petroleum Science and Engineering 45 (2004) 141–148

resistivity and density/resistivity overlay are 0.105 and 0.6, respectively. After calculating TOC amount using DLogR technique through three methods, namely, the sonic/resistivity, density/resistivity, and neutron/ resistivity, the sonic/resistivity method showed minimum amount of mean square error compared to the results from experimental sample analysis. Therefore, it can be concluded that this method may be useful in this well only, and it may not hold good in other wells. During borehole logging, for reducing the error obtained due to bad sonde,

145

it is better to apply average results of three DLogR methods as mentioned above. Based on this idea, the resulted values from the three methods were calculated, averaged and scaled with TOC amount from geochemical analysis of samples. So, the mean square error was equal to 0.18 in well bAQ. In addition, this technique was applied also for well bB.Q The surface distance between two wells is about 11 km. The mean square error was equal to 0.25, 0.2 and 0.25 for sonic/resistivity, neutron/ resistivity and density/resistivity logs overlay, respec-

Fig. 3. Sonic/resistivity overlay and calculated TOC (wt.%).

146

M.R. Kamali, A.A. Mirshady / Journal of Petroleum Science and Engineering 45 (2004) 141–148

Fig. 4. The comparison between TOC obtained from DLogR (sonic, resistivity) and TOC measured from core.

tively. In the three methods, mean square error was averaged as 0.15.

6. Neuro Fuzzy technique The applied mathematical method is Neuro Fuzzy, and Matlab software was used. Four logs including resistivity, neutron, sonic and density were considered as input data and TOC as output of the software (Fig. 5). Due to the lack of enough samples, in training of the Neuro Fuzzy technique, TOC data, obtained from DLogR technique, were

used. In well bA,Q a total of 500 samples were considered for training. Then, with no overlapping between training and test samples, in the testing section of the software, 268 samples were examined from shale lithology (Fig. 6) and 207 samples from carbonate lithology sequences. Average testing error in shale lithology is equal to 0.14, and in carbonate lithology, it is 0.15. In well QB,Q 451 samples were selected for testing. Thereby, the average test error is equal to 0.49. This error value is not desirable because TOC training values were selected from well bA.Q In this way, the software calculated TOC values pursuant to applying well bAQ (LOM=11),

M.R. Kamali, A.A. Mirshady / Journal of Petroleum Science and Engineering 45 (2004) 141–148

147

Fig. 5. Neuro Fuzzy diagram with four inputs (LLD, FDC, CNL, BHC) and one output (TOC).

while the value of LOM is equal to 10 for well bBQ.

7. Conclusion In order to determine the total organic carbon, two techniques have been applied in two pilot wells bAQ

and bB,Q through Pabdeh and Gurpi formations in the Dezful Embayment. The first used is the DLogR technique, which includes three methods using sonic/ resistivity, neutron/resistivity and density/resistivity logs overlay. TOC amounts were calculated based on experimental formula and then scaled with geochemical analysis data. The second technique is Neuro Fuzzy. Calculated mean square error values

Fig. 6. Showing test data and FIG out in well no. bAQ with shale lithology.

148

M.R. Kamali, A.A. Mirshady / Journal of Petroleum Science and Engineering 45 (2004) 141–148

using the DLogR technique for well bAQ and bBQ are 0.18 and 0.15, respectively, whereas the corresponding values for average test error obtained from Neuro Fuzzy technique are 0.14 for shale lithology and 0.15 for carbonate lithology in well bAQ and 0.49 in well bBQ. It seems logical to use Neuro Fuzzy technique for rich intervals through the same well or adjacent wells and fields where the geothermal gradient remains with no or little changes.

8. Recommendations (1)

(2)

(3)

(4)

Because DLogR is a simple and quick method in the recognition of the petroleum source rocks, it is recommended to use this method prior to the sampling of the intervals, where organic-rich layers are encountered. It is better to use logs such as natural gamma-ray tools (NGT) that determine radioactive properties of sediments individually. The radioactive elements generally associated with petroleum source rocks are thorium, potassium and uranium that have a direct relationship with organic richness in sediments. In order to increase epoch in the Neuro Fuzzy method, which results in minimum error, it is recommended to use computers with high RAM capacities. It is recommend to use DLogR technique in the same well and adjacent wells as well, but Neuro Fuzzy technique seems to be more applicable in the same well including other intervals.

References Beers, R.F., 1945. Radioactivity and organic content of some Paleozoic shales. AAPG Bulletin 26, 1 – 22. Dellenbach, J., Espitalie, J., Lebreton, F., 1983. Source Rock Logging: Transactions of 8th European SPWLA Symposium, paper D. Espitalie, J., Deroo, G., Marquis, F., 1985. Rock Eval Pyrolysis and Its Application (Reprints). Institute Francais Du Petrol. Geologie No. 207296, project B41 79008, 72p.

Fertle, H., 1988. Total organic carbon content determined from well logs. SPE Formation Evaluation 15612, 407 – 419. Herron, S.L., 1988. Source rock evaluation using geochemical information from wireline logs and cores (abs). AAPG Bulletin 72, 1007. Hertzog, R., Colson, L., Seeman, B., O’Brian, M., Scott, H., et al., 1989. Geochemical logging with spectrometry tools. SPE Formation Evaluation 4, 153 – 162. Hood, A., Gutjahr, C.C.M., Heacock, R.L., 1975. Organic metamorphism and the generation of petroleum. AAPG Bulletin 59, 986 – 996. Hunt, J.M., 1996. Petroleum Geochemistry and Geology, 2nd ed. W.H. Freeman and Company, New York. Hunt, J.M., Jaieson, G.W., 1956. Oil and organic matter in source rock of petroleum. AAPG Bulletin 40, 477 – 488. Hussain, F.A., 1987. Source rock identification in the state of Kuwait using wireline logs. SPE 15747, 477 – 488. Langford, F.F., 1990. Interpreting Rock-Eval pyrolysis data using graphs of pyrolizable hydrocarbon vs. total organic carbon. AAPG Bulletin 74, 799 – 804. Luffel, D.L., 1992. Evaluation of Devonian shale with new core and log analysis methods. SPE 21297, 1192 – 1197. Mendelson, J.D., Toksoz, M.N., 1985. Source rock characterization using multivariate analysis of log data. Transactions of the Twenty-Sixth SPWLA Annual Logging Symposium, paper UU. Meyer, B.L., Nederlof, M.H., 1984. Identification of source rocks on wireline logs by density/resistivity and sonic transit time/ resistivity cross plots. AAPG Bulletin 68, 121 – 129. Morel, J.A., 1999. Use resistivity as indicator of source rock maturity. Oil and Gas Journal, 72 – 74 (May 10). Passey, O.R., Moretti, F.U., Stroud, J.D., 1990. A practical modal for organic richness from porosity and resistivity logs. AAPG Bulletin 74, 1777 – 1794. Peters, K.E., 1986. Guidelines for evaluating petroleum source rock using programmed pyrolysis. AAPG Bulletin 70, 318 – 329. Schmoker, J.W., 1981. Determination of organic-matter content of Appalachian Devonian shales from gamma-ray logs. AAPG Bulletin 65, 2165 – 2174. Schmoker, J.W., Hester, T.C., 1983. Organic carbon in Bakken Formation, United States portion of Williston Basin. AAPG Bulletin 67, 2165 – 2174. Serra, O., 1986. Fundamentals of Well-Log Interpretation. The Acquisition Logging Data vol. 1. Elsevier. 679 p. Swanson, V.E., 1960, Oil yield and uranium content of black shales: USGS professional paper 356-A, p. 1–44. Tavish, M.C.R., 1998. Applying wireline logs to estimate source rock maturity. Oil and Gas Journal, 76 – 79 (Feb. 16). Tissot, B.P., Welte, D.H., 1984. Petroleum Formation and Occurrence. Springer-Verlag, New York. 966 pp. Tommi, O., 1994. Neuro-Fuzzy Systems in Control. Tampere University of Technology.