Application of seismic attributes and neural network for sand ...

Bulletin of the Geological Society of Malaysia 54 (2008) 115 – 121, doi: 10.7186/bgsm2008018Application of seismic attributes and neural network for sandprobability prediction — A case study in the North Malay BasinJi a n y o n g Ho u 1 *, To s h i h i r o Ta k a h a s h i 1 , Ar at a Kat o h 1 , Su w it Ja r o o n s i t h a 2 ,Kh u n Pu va n at Ch u m s e n a 2 & Ka z u o Na k aya m a 31JGI, Inc., 1-5-21 Otsuka Bunkyo-ku, Tokyo, 112-0012, Japan*Tel: +81-3-5978-8038, Fax: +81-3-5978-8050, E-mail address: hou@jgi.co.jp2CPOC, PETRONAS Twin Towers, Kuala Lumpur City Centre, 50088 Kuala Lumpur, Malaysia3JAPEX Co, Ltd., 1-7-12, Marunouchi Chiyoda-ku, Tokyo, 100-0005, JapanAbstract— The application of seismic data has greatly increased our capability to correctly understand the distributionsof reservoir layers and geological structures of interest. Seismic attributes analysis is a popular and important methodto predict the distributions of reservoir properties such as lithology, porosity and thickness. In the North Malay Basin,some sandstone layers related to reservoirs are found between A and B horizons. Especially, the gas reservoirs have beenconfirmed in the sandstone layers around Z horizon developed between A and B. Although 3D seismic survey has beenexecuted, the distributions of the sandstone layers, e.g. thickness, porosity, still remain unclear across the area becausethere are only few wells drilled. To understand the distributions of sandstone layers and lithologic change across the whole3D seismic survey area, we utilized Artificial Neural Network (ANN) method supported by Geology Driven Integration(GDI) technique. This process established the relationship between sandstone lithology and some seismic attributes. Weselected Z horizon as calculating datum plane to predict the sandstone distribution and lithologic change within the rangefrom 150 ms above the horizon to 120 ms below it, that is from A horizon to B horizon.Keywords: Artificial Neural Network, seismic attributes, rock properties, pseudo well, relationshipIntroductionWell logs and seismic records can be thought asresponses to change of material properties such as porosity,lithology, fluid content and bed thickness in subsurfacerocks. In general, the most effective and reliable methodto obtain the rock properties is well log analysis. Becausewell log is 1D data, however, the analysis can only obtainthe properties of underground rock or stratigraphy at welllocation and its neighborhood. Therefore, when only a fewwell log data are available (it is usual in the early stage ofexploration) or when the wells concentrate in a part of studyarea, it is usually difficult to understand the distributions ofrock properties across the whole study area. On the otherhand, each sample in a seismic trace has given amplitude,frequency, phase, etc. and seismic records, 2D or 3D, containusually very abundant information of lateral and verticalchange of rock properties. Therefore, seismic records areclosely related to well logs. If a mathematical relationshipcan been established between some seismic attributesfrom seismic records and rock properties from well logs,we can apply the relationship to predict the distributionsof rock properties away from well control in 2D or 3Dseismic survey area. The Artificial Neural Network (ANN)method supported by the Geology Driven Integration (GDI)technique provides such a tool for us.GDI concept and theoryThe GDI is a technique for predicting distribution andchange of rock properties, e.g. thickness, porosity, etc.directly from the seismic attributes (Nakayama & Hou,2002, Nakayama et al., 2003).In GDI, ANN method is utilized to establish themathematical relationship. A problem, however, is to collecta large number of well data to establish a statisticallysignificant relationship. Only a few available well datain the early stage of exploration had prevented us fromestablishing the relationship. To compensate such a conditionunder a few wells, a large number of pseudo wells can begenerated by a method of Monte Carlo Simulation in GDI(de Groot et al., 1996).The Monte Carlo Simulation is a procedure that involvessampling based on probability to approximate the solutionof mathematical or physical problems in a statistical way.There are two major advantages in this method superior toconventional ways. The first advantage is that we can steerthe algorithm with rules based on geological reasoning. Thesecond is that we can include hard constraints for each ofthe stochastic variables. In GDI, the Monte Carlo methodis used to generate pseudo wells based on the stratigraphicgeologicalmodel built from factual well data, geologicalinformation and geological knowledge. These pseudo wells

Ap p l i c at i o n o f s e i s m ic at t r i bu t e s a n d n e u r a l n e t w o r k f o r s a n d p ro b a b i l it y p r e d ic t i o ncontrolled by the movement of normal faults, both pureextensional and transpressional tectonics. The extensionaltectonics during Oligocene and Early Miocene causedtilted fault blocks, horst and graben features and anticlinalfeatures against faults (Figure 2). The trap style is mainlyfault dependent while widespread marine shales provide agood regional top seal. The intraformational shale in theprospective sequence also provides a very good top sealfor individual reservoir package. In the study area, somesandstone layers with reservoirs property are found betweenA and B horizons, and main reservoir is Z-sand betweenthem (Figure 3).Application of GDIUsually, the lithology information can be obtained fromlog data. In our study area, however, it is difficult to get itbecause there are six available wells and only three amongthem can be used in the study. Therefore, for understandingthe distributions of the sandstone layers between A and Bhorizons, the GDI technique is applied in the study area.a. Pseudo wellsTo predict the distribution of the sandstone usingthe seismic attributes, it is necessary to establish themathematical correlation, or relationship, between seismicattributes and lithologic property. Obviously, it is difficultto establish the statistical relationship only by limited threefactual well data. To obtain enough lithologic samples,we need more well data. For this reason, a 1D simulationcalled the Monte Carlo Simulation is introduced to generatepseudo wells to compensate the lack of factual well data(de Groot et al., 1996).A geological model is first established by integratingvarious real geological data including well logs, geologicalinformation and knowledge of the study area (Figure 4)before generating pseudo wells. In the geological model, thestratigraphic order, lithologic units and their properties inseveral formations are defined and are attached with variouslog responses, and various lithologies are distinguishedfrom relative GR value. For sampling the problem, we onlydefine two kinds of lithologic units in every formation unit,or sandstone and shale. Moreover, the sand and shale unitsare given a lothologic code respectively to distinguish andclassify them with seismic attributes. That is, sand is definedas one (1) and shale as zero (0).To obtain enough well realizations, 200 pseudowells with varying stratigraphic frameworks and well logresponses are generated (Figure 5 left). In this simulation,the parameters of variables such as thickness of stratigraphicunits, porosity and various log responses of lithologic unitsare changed simultaneously and stochastically, based onthe statistics of the factual well data and given geologicalconstraints. Such pseudo wells have various log responsesand are considered as similar to geological realities of thestudy area, and therefore, can be used as known samples oflitholgic properties, but they have not location information.Only purpose to generate pseudo wells is to compensate thelack of factual wells. Therefore, the location information isnot necessary in the simulation.The synthetic seismograms corresponding to the pseudowells can be constructed by convolving the reflectioncoefficient series with a zero-phase wavelet extracted fromthe real 3D seismic records (Figure 5 right). The pseudowells are an approach to lithologic properties of the studyarea, while the synthetic seismograms represent the featureof seismic attributes.b. Establishment of the mathematicalrelationship, utilizing ANNAs stated before, both well logs and seismic attributesmay be considered as a response to various rock properties.We can obtain the rock properties, such as lithology,thickness, porosity, fluid-bearing character, etc. from factualand/or pseudo well logs, and seismic attributes from seismicrecords or from synthetic traces. Since the relationship isnot necessarily linear, the ANN method that recognizesnon-linear correlation is used in GDI analysis.We train the ANN by feeding some known samplesof lithology and seismic attributes extracted from pseudowells and synthetic traces respectively into the ANN. Bythe training, the ANN can find a non-linear correlationGeological Society of Malaysia, Bulletin 54, November 2008Figure 3: Well correlation amongwells in the study area. Main sandreservoirs are found around Zhorizon.117

Ji a n yo n g Ho u, To s h ih i r o Ta k a h a s h i, Ar ata Kat o h , Su w it Ja r o o n s it h a, Kh u n Pu va n at Ch u m s e n a & Ka z u o Na k aya m aFigure 4: Geological-stratigraphic model established from realgeological information including well logs, stratigraphic sequence,lithologic units and geological knowledge.between lithology and seismic attributes. In this study,used seismic attributes, called “Sample”, is a time seriesof amplitude values along a given seismic trace. In otherwords, it returns the amplitudes at the specified samplingpositions within a specified time windows (Figure 6). When“Sample” is extracted, the extracting sample rate, the timewindow length at every sampling and total sampling timewindow length are necessarily specified. In this case, thesampling rate is 2ms and the time window length at everysampling is defined from -20 ms to 20 ms referring tosampling position, while the total sampling time windowlength is from -150 ms to 120 ms referring to Z horizon(Figure 6), which corresponds approximately from A horizonto B horizon (Figure 3).There are several ANN models, but the Multi-layersperceptron model with the learning algorithm of backpropagationis used to establish the relationship betweenthe lithology codes and seismic attributes (Figure 7). Theadvantage of the model is that it can feed back the errorinformation to input layer by comparing the calculated values(lithology property) by ANN with “actual” values extractedfrom wells (including factual and/or pseudo wells), and thenadjust automatically the weighting vectors interconnectingFigure 5: Part of pseudo wells (left)generated by Monte Carlo simulation andcorresponding synthetic traces (right), basedon the geological-stratigraphic model.Figure 6: Lithology extraction from pseudowells and seismic attributes (Sample orAmplitude) from synthetic traces as knownsamples.118Geological Society of Malaysia, Bulletin 54, November 2008

Ap p l i c at i o n o f s e i s m ic at t r i bu t e s a n d n e u r a l n e t w o r k f o r s a n d p ro b a b i l it y p r e d ic t i o nFigure 7: The trained ANN (left) and its performances (right)for establishing the relationship between the lithologic code andseismic attributes. Nodes are displayed in color scale relative totheir contribution to the output, i.e., from red (highest) , orange,yellow to white (lowest).Figure 8: A lithology cube created by applying the ANN to 3Dseismic volume. It shows the distribution of sand probability fromA horizon to B horizon within the seismic volume.Figure 9: The distribution of sand probability on thehorizon slices with different time.Geological Society of Malaysia, Bulletin 54, November 2008119

Ji a n yo n g Ho u, To s h ih i r o Ta k a h a s h i, Ar ata Kat o h , Su w it Ja r o o n s it h a, Kh u n Pu va n at Ch u m s e n a & Ka z u o Na k aya m aadjacent layer so that the RMS errors can be decreased to aacceptable or reasonable level. According to the experiences,a RMS error below 0.8 is considered generally to bereasonable. The right lower figure in the Figure 7 showsthe decay of the calculative normalized RMS error duringtraining. The final error is about 0.8. Although it is a littlehigh, it is still thought in a reasonable range.Figure 10: Sand distribution on the Z horizon slice.c. Prediction of distribution of sand probabilityThe trained ANN is applied to a 3D seismic volumein the North Malay Basin to predict the distribution ofsand probability within the interval about from A horizonto B horizon.The Figure 8 is a lithology cube showing the distributionof sand probability in the specified interval of the 3D seismicsurvey. It is created by applying the trained ANN above to the3D seismic volume of the North Malay Basin. The Figure 9shows the distribution of sand probability on several horizonslices with different depth, which is generated parallel to theA horizon and each 20ms interval. From these illustrations onthe Figure 9, we can find there are little sand probability onsome horizon slices, e.g. A horizon+20 ms, A horizon+120ms and A horizon+160 ms, etc. This feature suggests somepredominant zones of shale or argillaceous rock around thesehorizon slices. Moreover, on the Z horizon slice (Figure10), the distribution of sand probability shows a channeland fan in shape. This feature may indicate the depositenvironment of sandstone. On the lithology sections alongInline or Crossline (Figure 8), the most of the high sandprobability zones (red color) are observed in discontinuousdistributions. This may indicate that sandstone develops indiscontinuous sand body or lens.ConclusionsFigure 11: Comparison of predicted lithology by GDI with GRfrom factual well 1 at well locations.(Solid line: sand probability,Dotted line: GR).Figure 12: Comparison of predicted lithology by GDI with GRfrom factual well 2 at well locations.(Solid line: sand probability,Dotted line: GR).120The Geology Driven Integration Tool including theartificial neural network and Monte Carlo simulationtechniques is applied to the 3D seismic survey of the NorthMalay Basin to predict the distribution of sand probabilitywithin the main reservoir horizons. The predicted resultis checked with GR from factual wells (Figures 11a and11b). The check shows that the predicted result is consistentwith log features on the whole although there is error tosome extent.Using the seismic attributes and ANN technique isan effective method in the prediction of distributions ofrock properties, especially when factual well data are fewor too concentrated so that it is difficult to understand thedistributions by conventional methods.AcknowledgementsWe thank Carigali-PTTEP Operating Company Sdn.Bhd (CPOC) and Malaysia-Thailand Joint Authority (MTJA)for permission to publish this paper, which is based on thedata and information of seismic and geology from a fieldsituated in the North Malay Basin.Geological Society of Malaysia, Bulletin 54, November 2008

Ap p l i c at i o n o f s e i s m ic at t r i bu t e s a n d n e u r a l n e t w o r k f o r s a n d p ro b a b i l it y p r e d ic t i o nReferencesDe Groot, P.F.M.,. Bril, A.H., Floris, F.J.T. and Campbell, A.E.1996, Monte Carlo simulation of wells, Geophysics, 61(3),631-638.Nakayama, K. and Hou, J., 2002, Prediction of ReservoirProperties by Monte Carlo Simulation and Artificial NeuralNetwork in the Exploration Stage. Soft Computing forReservoir Characterization and Modeling. Edited by P.Wong,F.Aminzadeh and M.Nikravesh, pp. 15-33.Nakayama, K., Hou, J., Ohkawa, S. and Mitsuda, S. 2003, Applicationof Geology Driven Integration Technique in the Prediction ofReservoir Distribution: An Example from the Japan Sea. Thepaper collection of the 9th Formation Evaluation Symposiumof Japan, SPWLA.Rumelhart, D.E., Hinton, G.E. & Williams, R.J., 1986. Learninginternal representation by error propagation. In: Rumelhart, D.E. & McClelland, J. L. (eds.) Parallel Distributed Processing.Cambridge, MIT Press, pp. 318-362.Manuscript received 28 May 2008Revised manuscript received 25 November 2008Geological Society of Malaysia, Bulletin 54, November 2008121