Gerencia.pdf

SPE-191695-18RPTC-MS Complex Approach to Fault Description While Geosteering for Maximization Reservoir Contact in Horizontal Wells in West Siberia Oilfields Vladislav Vadimovich Krutko, Tatyana Alexandrovna Yurkina, and Dmitriy Yurievich Kushnir, Baker Hughes; Valery Borisovich Karpov, RITEK Copyright 2018, Society of Petroleum Engineers This paper was prepared for presentation at the SPE Russian Petroleum Technology Conference held in Moscow, Russia, 15–17 October 2018. 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 paper presents an interdisciplinary approach to the tectonic dislocations description based on the results of interpretation of seismic data, petrophysical analysis of well-logging data in horizontal wells and the results of inversion of the multifrequency propagation tool. On the materials of seismic surveys and loggingwhile-drilling data in horizontal wells of one of the oil fields in Western Siberia a cinsistent approach to fault identification and description is presented. Interpretation of seismic data allows predicting the fault presence along the horizontal well profile and gives preliminary estimation of the dipping and amplitude parameters of the fault surface. Utilizing the the logging-while-drilling data of a group of horizontal wells with the aim to improvement the geological model after fault crossing an approach including machine learning principles is proposed and tested. Application of inversion of the multifrequency propagation tool to one of the updated geological model with faults allowed to improve the estimations of the fault zone parameters.

Introduction The quality drop of the residual oil reserves due to increased intensity of the oil production have been taking place over the world. The part of the hard-to-recover reserves in those reservoirs characterized by complex geological conditions (low-pearmeable and thin reservoirs, bottom water-drive and gas deposits with oil rims, with complex influence of disjunctive structures) continuously increases. For these objects with hardto-recover reserves it is necessary to actively apply the procedures focused on the sweep efficiency increase: well network density increase, horizontal wells, side-tracks drilling, fracturing, hydrodynamic methods, flow deviation methods (Lutfullin 2009). Horizontal wells have become one of the most efficient methods for increasing of producing area while approaching the following tasks (Molchanov et al. 2001): – Well productivity raise due to filtration area increase; – Water-free well production period prolongation; – Gas breakthrough probability inhibition by the means of wellbore depression of reservoirs decrease;

2

SPE-191695-18RPTC-MS

– Involvement in the development reservoirs with low poroperm properties or highly viscous oil; – Development of the hard-to-reach oil and gas fields. For the successful drilling the horizontal wells the geosteering is necessary that is positioning and correction of the wellbore direction (Beer et.al. 2010). Based on information received during the drilling the distance from wellbore to top and bottom of the reservoir is estimated as well as to gas-oil-water contact and shale layers. The goal of the geosteering is optimization of the horizontal well trajectory so that the productivity of the drilled well is at its maximum. Timely steered drilling trajectory allows, in particular, prevent water-bearing formation or gas cap penetration as well as to place the most optimal position of the wellbore in pay zone. Before the horizontal well drilling usually a pre-model of the part of the target reservoir that is supposed to be penetrated is created. In this model the available information about the properties of the top and bottom strata as well as lateral variability of the properties is summarized. For this the core data and well log data from the nearby vertical, deviated and pilot offset wells are used. As a rule, the properties estimated in offset wells. During the drilling of the horizontal section while new data receiving from logging-while-drilling (LWD) tools the geological model is updated and corrected until matching of the synthetic and measured logging data. That allows to specify structural model of the target reservoir and get the information for new wells planning. For the decision making while geosteering the drilling parameters (rate of penetration, rotation rate of bit, load on bit etc.), mud and those data coming from LWD tools are used. The LWD data allows to perform realtime estimation of the lithology and poroperm properties and correlation with offset wells. Very interesting and promising methods while geosteering are azimuthal images (gamma, density or resistivity) which allow to identify and estimate dips and azimuth of bedding surfaces. Usage of azimuthal images allows to identify position of tectonic disjunctions, such as faults, which increases effectiveness of the geosteering while drilling. In addition to images the azimuthal low- and high frequency multiple propagation tools showed its effectiveness in mapping reservoir borders up to tens of meters. It is reached by combination of multiple transducing and receiving coils in the tool induction coils in usage with specialized algorithms of multi-layered inversion (Tilsley-Baker et al., 2016). Equipment of the drillstring with such azimuthal electromagnetic tools allows to perform so called pro-active geosteering that is allows to make decisions well trajectory deviation before penetration layers of different properties with bit. However last tendency of oil companies is to reduce the well-logging program in horizontal wells up to gamma and resistivity methods in order to optimize costs while building wells. This considerably cuts the volume of available information about the reservoirs. In this situation mud logging analysis gives invaluable and robust information about lithology and mineralogy. During stratigraphic analysis based on standard well-logging methods it is not always possible to differentiate top and bottom clay seals that isolate the target reservoir from gamma, neutron, resistivity and density measurements. This leads to misunderstanding or wrong interpretation of the reservoir structure, especially in reservoirs affected by faults. Here it can be helpful to analyze the results of mud logging that allow to identify distinguishable layers of different lithology. Even more valuable results can be received when using automatized mineralogical analysis of geological samples, widely used for mineralogy estimation in industry from the beginning of the technology since 1982's (Miller et al, 1982). Another modern tendency in is to use seismic survey data for horizontal well geosteering. Usage of seismic data and applying of the results of its' interpretation in combination with other measurements (LWD, mud logging) give an opportunity to confidently drill horizontal sections of wells, apply early corrections to trajectory deviation, increase effective net-to-gross that leads to decrease of construction time and costs. In the paper the possibilities of modern methodological and technological approaches usage are shown that allow perform the optimal positioning of the horizontal wellbores in pay zone of target reservoir

SPE-191695-18RPTC-MS

3

in the environment of the tectonic disjunctions as sub vertical faults. The experience on step-by-step specification of the reservoir structure in close vicinity to the fault based on seismic interpretation, welllog data qualitative and petrophysical quantitative interpretations and lithotyping is summarized. Special attention should be put on necessity of the usage of the seismic data on preliminary horizontal well profile modelling as well as petrophysical analysis in order to distinguish lithotypes for geonavigation during drilling. The approach described can be interesting to those specialists in the area of planning and horizontal well construction. The integrated approach to geological support of the drilling process and minimizing risks of the pay zone loss can be used, optimized and improves on any developed oil filed.

Seismic methods of fault tectonics interpretation The estimation of seismic methods of fault detection was performed on the materials acquired on one of the oil filed in Frolov oil-and-gas district in Western Siberia, Russia, RITEK oil company. The observed territory of the oil filed is characterized by complex geological structure, namely not high effective reservoir thickness (13-24 m, 18 m average), thin layering of sandstones and silts with effective thickness of reservoirs 11.4 m, the presence of low porosity, low permeability zones of reservoir and tectonic block structure of the reservoir. Thickness of layers does not exceed 4.0 m, mostly occurred layers are of 0.4-1.0 m, with lithological stratification in 2-6 sub-layers. Porosity type of reservoirs rocks – integranular, with porosity varying in range 13.2 – 24.1%. Weighted average of porosity in the main deposit is 16.6%, permeability is low, 0.5 mD in average. Oil deposit of the AC3 reservoir horizon is lithologically sealed. Water-bearing reservoirs are not presented in the offset vertical wells. Productivity of the reservoirs is not high, oil rates are from 0.2 m3/d to 8.1 m3/d initially, increased in 1.4-17 times after intensification. While working with hard-to-recover reserves it is important to have geological model with the identified zones of high risks to meet a fault. Drilling in reservoirs of low thickness it is critically important to know precise position of the horizontal wellbore relative to structure avoiding coming out of reservoirs. The base for the drilling planning is a structural map of the reservoir top. After the field jobs and receiving the results of the wide-aperture three-dimensional seismic survey the considerable refinement of the top structure of the reservoir (Fig. 1a) and the morphology of all the earlier identified structural forms was done. A faulted and blocked principal model of the reservoir was created which formed the basis for well pads positioning scheme on the oilfield territory, as well as for usage in horizontal wells drilling.

4

SPE-191695-18RPTC-MS

Figure 1—Structural map in the area of the of the 2301G well (a) and initial geological model (b).

Following an example of successful real-time usage of the 3D seismic data analysis for the horizontal well profile correction. Horizontal well 2301G was planned in north-west. For the reservoir model construction, a structural map prepared by Baker Hughes specialists from the seismic survey interpretation was used. In the 2301G well drilling direction in the first third of the horizontal section sub horizontal bedding of structure was expected. In the middle and last third of horizontal section small rise angle from 0° до 0.3°. An outcrop of the structural map with the well 2301G place is presented on Fig. 1a and the initial geological model is presented Fig. 1b. Rare well network in the area of the drilling led to high uncertainties in structure parameters and in risk to penetrate a fault. As an offset the well 75B3 was used (Fig. 1a). In the offset well 75B3 the horizon AC3 represents thin layering of silt, shaly sandstone and tight rocks. Thickness of the AC3 horizon is 13.2 m. In the Fig. 2 the typical stratigraphic cross-section of the AC3 horizon in the well 75B3 is presented.

SPE-191695-18RPTC-MS

5

Figure 2—Stratigraphic cross-section of the AC3 horizon in the 75B3 offset well.

During the drilling of the transport section the wellbore of the 2301G well penetrated the top of the AC3 horizon at the – 1912.12 m TVDSS that completely coincided with the structural surface prognosis based on seismic data. Error of the top depth estimation did not exceed 0.13 m. Fig. 3 shows the structural decomposition attribute cross-section (a) and the map of the dip of maximum similarity attribute (DOMS) at the top of AC3 horizon (b) that have been used for fault detection and tracing.

Figure 3—Cross-section of the spectral decomposition (SD) attribute along the 2301G well profile (a) and the map of the DOMS attribute at the top of the AC3 horizon

The blocked model of the AC3 horizon is complex and the vision of the structure of the target reservoir can change significantly while drilling and penetrating the reservoir. According to seismic data there is possibility to cross three faults along the planned horizontal section of the 2301G well.

6

SPE-191695-18RPTC-MS

Figure 4—Initial geological model with the fault included, well 2301G.

After penetration of the top of the target horizon the well profile was stabilized at 90 degrees within the AC3. At the measured depth 2883.5 m an abrupt increase of gamma measurements that was interpreted as fault crossing and coming out to silt. The wellbore crossed fault of the throw type with amplitude of approximately 5.5 – 6.0 m. Based on mud report the shale content increase from 5% to 15%. Logging measurements: GR ~105-115 gAPI, Res ~ 7-8 Ohm.m, Density ~ 2.52-2.57 g/cc. Mud report showed sandstone ~70%, silt ~15%, shale ~15%, argillite ~10% (Fig. 5). Based on experience of the previous well drilling it was known that in this oil filed the top shales have different composition from the bottom shales according to mud analysis. The top shales above the AC3 horizon consist mostly of argillites (up to 85%), whereas the bottom sediments consist mostly of sandstones and silts with gradual increase in argillite from 15% of argillites in reservoir to 85-90% in the interval 30-50 m lower the bottom of the reservoir. From the current drill depth, the well deviation drops to 89 degrees was performed following the stabilization. From the measured depth 2980.0 (1917.0 m TVDSS) the drop of the gamma ray to 75 gAPI and the raise of the resistivity up to 13-15 Ohm.m were observed, the wellbore returned to the target horizon. The forecasted from the seismic tectonic disjunction of the type "throw" was met at the measured depth 3459.0 m (amplitude ~ 15 m), after which the wellbore penetrated the top argillites. According to the structural map either increase of the top structure or a fault of "thrust type" in the interval 3585.0 m ±50.0 m was expected, also interpreted from seismic data. A recommendation to continue to drill the horizontal section with stabilization at 90 degrees was proposed. The drilling was continued in argillites above the horizon. From the measured depth ~ 3610.0 m the return of the wellbore to the AC3 horizon was registered from top to bottom and drilling continued at the top of the horizon, supposedly, in the second tight layer. The measurements were GR ~60 gAPI, resistivity ~20 Ohm.m, density ~2.6 g/cm3, ROP ~15m/h. Mud logging: sandstone ~40%, argillite ~ 40%, silt ~10%, shales ~10%. (Fig. 6). At the measured depth 3658.0 – 3676.0 m the wellbore of the 2301G well crossed the second tight layer. From the measured depth 3676.0 m the wellbore came into target reservoir with gamma ~ 60-80 gAPI, resistivity 12-15 Ohm.m, density ~ 2.35 – 2.43 g/cc. Further, according to the geonavigation model the drilling was performed in the target reservoir below the tight layer. The drilling of the horizontal section was completed at final depth of 4289.0 m, the tasks offered t the geonavigation team were accomplished. Lithological and stratigraphic composition of the AC3 horizon was confirmed. The information about the horizon top structure behavior, type a deposition of the tectonic disjunctions or faults, received while analyzing seismic data allowed to keep slightly sloping trajectory of the well and establish its optimal drilling in complex structural and tectonic environment. Thus, actual analysis of available seismic data allowed making correct decision about horizon position relatively to the met faults.

SPE-191695-18RPTC-MS

Figure 5—Geological model corrected based on actual data after crossing the first fault, well 2301G.

Figure 6—Geological model corrected based on actual data after crossing the second fault, well 2301G.

7

8

SPE-191695-18RPTC-MS

Figure 7—The final geological cross-section along the 2301G well profile, combined with seismic section.

Statistical and inversion methods for fault description In this and next parts the multivariate statistics methods with their application for lithotyping and usage of this experience for fault model optimization based on electromagnetic well-logs inversion will be discussed. Multivariate Statistics in Lithofacial analysis The approach to the lithological description during interpretation of well-log data in horizontal wells is different from interpretation in slightly deviated and vertical wells. Besides the differences associated with the drilling environment and its effect on well logging measurements, such as flushed zone occurrence in permeable formations near wellbore and its affect on logging measurements (Eltsov, I. et al, 2011), anomalous resistivity reading from electromagnetic propagation tools in horizontal wells while crossing layer boundaries (Sviridov et al., 2014), the main difference in well logs interpretation in horizontal sections is the approach how lithotyping and stratigraphic breakdown along the horizontal borehole are performed. In horizontal wells petrophysical analysis of the crossed formations has to be extended and complemented with methods of geonavigation solving the problem of positioning of the wellbore and should be based on qualitative and quantitative identification of penetrated lithotypes from a variety of well-logging methods and structural construction during the drilling. In vertical wells lithological breakdown is performed under assumption that the layers that form some order in vertical direction keep this order along the measured depth in borehole. However, this rule of stratigraphic ordering may not be kept in deviated and horizontal wells. The main reason for crossed lithotypes not to follow each other along the horizontal borehole is highly ambiguous and unpredictable

SPE-191695-18RPTC-MS

9

lateral changes of rock types. Description of lateral rock changes is a complicated task due to the following (but limited by) reasons: – mechanisms of sedimentation; – post-depositional processes (wreathing and erosion); – relative shifts and transport of rock masses due to tectonics. Correct lithotyping allows, besides justified application of petrophysical models for poroperm properties estimation, also apply the results of lithotyping for horizontal wellbore positioning after crossing a fault, with further optimization the geonavigation model. Some examples from the same formation AC3 will be shown. The part of the AC3 formation associated with pad 16 is characterized by heavy compartmentalization due active tectonics in past. Only part of the faults was predicted by seismic data interpretation, considerable number of faults were detected only during the drilling basing on direct and indirect features. At the stage of well planning and decision making while geosteering in the faulted environment the ability to describe a fault in details basing on combination of the a-priory knowledge and results of the areal and borehole geophysical survey has big importance. Due to low spatial resolution and, often, due to low quality of the seismic data it is often only possible to flag existence and absence of the fault on seismic cross-section along the horizontal wellbore, whereas the amplitude and direction can be evaluated with big error or not at all. On the other hand, well-log data has higher resolution along the borehole, higher density of measurements and can describe the penetrated rock quantitatively. Litho-facies analysis method applied on well-log data allowed to more precisely describe the crossed rock in order to justify to position of the wellbore. On the pad 16 there were drilled 4 wells, 584G, 584G-BIS, 5841G and 5842G, horizontal sections of which have penetrated the sediments of the AC3 formation. Vertical offset wells 545, 583 and 76P, the data of which have been used for preliminary modelling for geosteering, were characterized by standard wireline well-logging measurements, so that only gamma ray and resistivity measurements in were presented in each well, whereas in the well 545, bides gamma ray and resistivity, the neutron data was available, in 583 only the acoustics and spontaneous polarization measurements, in 76P – neutron and density. Neither of offset wells was characterized with photoelectric factor measurements. So, for preliminary stratigraphic breakdown and correlation of AC3 formation layers for geonavigation was performed only basing on gamma ray and resistivity measurements. As it is seen from Fig. 8 the target formation AC3 is inhomogeneous, sandstone, siltstone, tight rocks and shales are presented. Based on proposed lithotyping from two measurements of gamma and resistivity the series of geonavigation scenarios were built for each of the horizontal well drilled on the pad 16. However, the such a build geonavigation model has serious drawback, that can be summarized as following: – stratigraphic breakdown based on gamma ray and resistivity only can be very rough due to low variability and low data density of these two measurements and so there is little confidence in ability further recognize the identified lithotype with LWD methods. – the usage of gamma ray and resistivity measurements in lithotyping and correlation between offset wells does not give any idea on how other well-logging methods utilized while drilling should behave; – identified and correlated lithotypes are "prolonged" along the structural surface of the AC3 formation making the geological model unrealistic.

10

SPE-191695-18RPTC-MS

Figure 8—Stratigraphic breakdown and correlation of the layers in AC3 formation between offset wells 545, 583, 76P.

During the drilling of the horizontal wells on the pad 16 the LWD logging tools of 155 mm diameter were used that allowed to measure and transmit in real-time the following measurements: natural gamma ray, density, photoelectric factor and thermal neutron porosity, as well as resistivity from multifrequency propagation tool. In this work an attempt to describe in bigger details the AC3 formation based on highly dense data of four basic LWD measurement from drilled horizontal wells was made without direct usage of the offset well data. Breakdown and lithotyping were done based on GR-Density-PE-Neutron measurement after drilling of all four wells 584G, 584G-BIS, 5841G and 5842G on the pad 16 and should include two steps. At the first step the so called electrofacies (Serra, 1985) are identified as such a linear or non-linear combinations of four logging measurements (gamma, density, neutron porosity and photoelectric factor) that form distinguishable groups of measurements that are similar to each other within their own groups and are different enough from the measurements in other groups. At the second step there is procedure creates statistically justified one-to-one mapping of the identified electrofiacies of four logging measurements with a-priori known lithotypes identified from mud logging or core data. From statistical point of view, the first step of the electrofiacies identification from logging measurements is referred to as clustering problem and is can be successfully solved with incorporation of unsupervised machine learning algorithms. The second step of one-to-one mapping of the identified electrofacies with known lithotypes is referred to as classification and can be solved with the supervised machine learning methods (Martinez et al, 2010). In the Fig. 9 the possible combinations of visualizations of four LWD measurements in the wells 584G, 584G-BIS, 5841G are shown. All the measurements are made in the AC3 formation. As it can be seen the paired visualization of the LWD measurements does not lead to evident clustering, leaving the whole set of data mostly as a single spot.

SPE-191695-18RPTC-MS

11

Figure 9—Visualization of pairs of four LWD measurements in all wells 584G, 584G-BIS, 5841G.

Clusters identification is performed, as a rule, in unsupervised way, i.e. without preliminary knowledge on how many clusters there should be in space of measured data variables, what their sizes and shapes are. This kind of uncertainty makes the cauterization problem highly ill-posed inverse problem in classical sense, which shows non-unique of solution in terms of clustering and clustering can be unstable with small variation in input data. In order to reduce all or some of these uncertainties related to clusters identifications process it is good practice to perform some kind of dimensionality reduction of the multivariate data. Dimensionality reduction is a pre-clustering statistical procedure that allows to project multivariate dataset onto lower dimensional 3D or 2D spaces. The main and most widespread dimensionality reduction algorithms are Principal Component Analysis (PCA) and Factor Analysis (FA). These algorithms are extremely effective if the linear relationships between measurements can be assumed linear in some sense. In this work for four LWD measurements a non-linear algorithm of dimensionality reduction t-SNE (tdistributed stochastic neighbor embedding) (Van der Maaten, Hinton, 2008) has shown the most efficient natural clustering of the LWD data. The results of dimensionality reduction of the LWD measurements dataset are shown in Fig. 10. It is seen that the Principal Component Analysis (Fig. 10a) does not allow to identify visually that does not give basis to rely on auto cluster algorithms after PCA application. On the other hand, non-linear dimensionality reduction algorithm t-SNE shows much more versatile clustering (Fig. 10b). Such a natural segregation of the data in clusters on a two-dimensional surface give hope that first, the auto clustering will be much more robust, and second, this clustering in LWD measurements space can actually reflect the spatial segregation of the clusters in 3D space and there is a natural mapping the lithotype into identified clusters of the electrofacies. In the Fig. 12 the 3D profiles of all four wells are shown with associated coloring according to distinguished clusters.

12

SPE-191695-18RPTC-MS

Figure 10—Results of dimensionality reduction on the LWD measurement dataset GR-Density-PE-Neutron: a – Principal Component Analysis, b – t-SNE.

Figure 11—Manual clustering of the combined LWD measurements based on: a - t-SNE visualization, b – isolated clusters on PCA plane.

SPE-191695-18RPTC-MS

Figure 12—Spatial distribution of isolated electrofacies in the drilled part of the AC3 formation (a) and their occurrence in each well separately: 584G (b), 584G-BIS (c), 5841G (d), 5841G (e)

13

14

SPE-191695-18RPTC-MS

Figure 13—Well profile and well-logging data in the well 584G. At the measured depth 2673.2 m a fault of the "thrust" type was met.

Figure 14—Well profile and well-logging data in the well 584G - BIS. At the measured depth 2828.0 m a fault of the "thrust" type was met as in the 584G well.

SPE-191695-18RPTC-MS

15

Table 1—Lithotypes and electrofacies association based on drilling results, AC3 formation

Logging-While-Drilling Resistivity Data Inversion near the Faults As an example of the detailed analysis of the fault parameters combining lithofacies breakdown, images and electromagnetic inversion of the induction measurements in the AC3 formation the 5842G well was analyzed. In the Fig. 15 (a) an interval of the well profile is shown where the well crosses an assumed fault at MD 3635.0 m. The well profile after the drop of the inclination crosses in order similar by lithotype electrofacies of non-homogeneous sandstones (red-yellow) before and after the fault (Fig. 15b). However, based on detailed density image interpretation another fault at 3485.0 m was assumed (Fig 15c). Thus, based on direct fault detections from density image and indirect signs of lithotyping order from lithofacies analysis the final updated geological model with two faults was proposed (Fig. 15d).

16

SPE-191695-18RPTC-MS

Figure 15—Preliminary fault model (a) and final fault model (d) based on images lithofacies interpretation; well profile with lithotypes coloring (b) and LWD well logs (c).

All the wells in the pad were placed using logging-while-drilling deep resistivity measurements. The tool operates on 400 kHz and 2 MHz frequencies, measuring the phase difference and attenuation. These measurements make possible to estimate of the layer resistivity, in which the tool is located, and the resistivities of the adjacent layers. Inversion in the framework of 1D layered and 2D formation models is proposed to improve the formation geoelectrical model near the faults zones. Deep resistivity tools are sensitive up to 3-5 meters from the well (Sviridov et al., 2014). Due to the axial symmetry of the transmitter and receiver coils around the tool axis, measurements are devoid azimuthal sensitivity. Therefore, a priory information about lithological structure and layer resistivities is necessary to reduce geological uncertainty. An expected geoelectrical formation model is based on the preliminary petrophysical analysis of the data and the refined geonavigation reservoir model. This is the difference

SPE-191695-18RPTC-MS

17

compared to the approach given in the paper (Dupuis et al., 2014). Geoelectrical formation model is improving in two stages: the first is multi-parametric inversion in 1D layered formation model and the second is multi-parametric inversion in 2D formation model. Let us briefly describe the scheme of data interpretation within the framework of 1D layered model. An expected 1D layered reservoir model with specified number of layers is set up along a measured depth of a well at a certain interval. The positions of the boundaries between the layers relative to the tool, the resistivities of the layers and the dip angle of the formation are adjusted with ranges for all the parameters. For each parameter, restrictions are in accordance with a priori information. In our case, in the absence of the azimuthal sensitivity of the measurements, the constraints on the parameters are established from the refined geonavigation model. The inversion algorithm minimizes the objective functional, consisting of the sum of the residual between the synthetic and measured curves, the deviations between the recovered and expected parameters of the model, and the "penalty" functions responsible for the parameters constraints (Sviridov et al., 2014). The residual is calculated for the attenuation amplitudes and phase differences and is normalized to the standard tool error: the relative error is 2%; the absolute error is 0.01 dB for the attenuation amplitude and 0.1 degree for the phase difference. Simulation and inversion in 2D formation model is performed using an algorithm based on the method of boundary integral equations (Dyatlov et al., 2017). The 2D model is constructed on a certain interval starting from the set of 1D layered models at smaller intervals. According to the refined geonavigation model of the reservoir (Fig. 15d), the faults are located at depths of about 3485 and 3635 m along the measured depth. The lithological type of the formation layers, the order of layers and the resistivity ranges are considered known when constructing a series of 1D layered formation models. Fig. 16 shows 1D layered geoelectrical model in the vicinity of the first fault preliminary located at the measured depth of 3485 m. The 1D model is recovered as the result of interval-by-interval inversion with 2 m interval size in accordance with the refined geonavigation model.

Figure 16—1D layered geoelectrical model in the vicinity of the first fault in the interval 3480-3492. The resistivities of the layers and the distances to the boundaries are indicated.

18

SPE-191695-18RPTC-MS

Based on the1D layered model, an expected 2D formation model is constructed. According to the seismic data, the faults are sub vertical. In this case, the tool measurements are weakly sensitive to the fault dip angle. Therefore, in the expected 2D formation model, the fault is considered vertical and the position of the fault plane along the measured within the interval 3480-3492 is recovered. Small changes in the resistivities of the 2D model domains and positions of the boundaries between the domains are also allowed in the inversion process. Fig. 17 shows the resulting two-dimensional model.

Figure 17—The resulting 2D geoelectrical model and logging curves in the vicinity of the first fault in the interval 3480-3492. The fault is located at the measured depth of 3488 m.

As a result of the 2D modeling and inversion, the 2D formation model near the fault is recovered and the position of the fault plane is adjusted at the measured depth of 3488 m. Fig. 18 shows the measured and synthetic of the 1D and 2D models curves. The synthetic curves obtained in the 2D model are smoother and do not have sharp jumps. In the interval 3484.5-3485.5, in accordance with the measured data and the density image, the well trajectory intersects the subvertical high-resistance inclusion, which is considered and fitted in the 2D model. The fault throw amplitude is approximately 3.9 m.

SPE-191695-18RPTC-MS

19

Figure 18—The measured and synthetic logging curves on the interval 3480-3492: attenuation amplitude shown on the top, phase difference shown on the bottom; for frequency 400 kHz shown on the left, for frequency 2 MHz shown on the right. Measured data are indicated by the black dashed line with the gate of noise of the one standard tool error, 1D synthetic data are indicated by the blue solid line, 2D – by the red solid line.

By analogy with the first fault, the resistivity log curves are interpreted near the second fault at the interval 3630-3640. Fig. 19-20 show, respectively, the 1D layered model, the 2D model, and the measured/ synthetic curves. From the variation of the curves along the depth, it can be concluded that the contrast of the resistivity of the layers in which the tool is located before and after the fault is significant. The resistivities differ to a greater extent (9 and 18 Ohmm) than in the case of the first fault (19 and 15 Ohm.m. The position of the inflection point of the curves (measured depth of 3634.5 m) gives some estimate of the fault position. The 2D inversion yields a more accurate value of 3634 m. In the interval 3636-3637, as well as near the first fault, in accordance with the density azimuth image (GGC) and logging curves the well trajectory intersects the sub vertical high-resistance inclusion, which is also taken into account in the 2D model. The fault throw amplitude is approximately 1.5 m.

20

SPE-191695-18RPTC-MS

Figure 19—Layered geoelectrical model in the vicinity of the second fault in the interval 3630-3640. The other notations are the same as in Fig 16.

Figure 20—The resulting 2D geoelectrical model and logging curves in the vicinity of the second fault in the interval 3630-3640. The fault is located at the measured depth of 3634 m.

SPE-191695-18RPTC-MS

21

Figure 21—The measured and synthetic logging curves on the interval 3630-3640. The other notations are the same as in Fig. 18.

Conclusion A combined approach to the fault identification and description based on 3D seismic surveys and detailed analysis of the LWD well-logging data was observed. The proposed approach includes step-by-step retrieving information about faults existing in the target geological zone with following specification and adjustment of the fault parameters. The areal 3D seismic surveys as the main method used for fault identification and their continuity bring the biggest volume of information about structure of the faulted reservoir. However, due to limitations in spatial resolution of seismic data these data have the lowest accuracy in fault parameters, such amplitude and dip, estimation. Qualitative and quantitative interpretation of the LWD well-logging data give much higher vertical and lateral resolution of the formation inhomogeneity because of higher resolution along the wellbore and higher measurement density. In case of well logging in deviated and horizontal wells the interpretation of well-log data is complicated by lateral variability of certain lithotypes. It was shown that the machine learning methods for analysis of multivariate data from LWD well-logging tools, in particular, non-linear dimensionality reduction algorithms allow to efficiently visualize and cluster the measured data and perform the detailed stratigraphic breakdown. Lithofacies analysis along with traditional image analysis allowed to improve geonavigation model. The most accurate and, at the same time, the most demanding to computational resources and to the quality of initial geological model is two-dimensional electromagnetic inversion of the resistivity data near the fault surface. The inversion allows to model realistic signals of resistivity tools and estimate detailed parameters of dip and angle of the studied fault.

References

Beer, R., Dias, L., Cunha, A., Coutinho, M., Schmitt, G., Seydoux, J., Morriss, C., Legendre, E., Yang, J., Li, Q., 2010. Geosteering and/or Reservoir Characterization The Prowess Of New-Generation Lwd Tools: SPWLA 51st Annual Logging Symposium, June 19-23, Perth, Australia, 2010, SPWLA-2010-93320.

22

SPE-191695-18RPTC-MS

Dupuis, C., Omeragic, D., Chen, Y., Habashy, T., 2014. Inversion-Based Workflow to Image Faults Crossed by the Wellbore Using Deep Directional Resistivity Provides New Way of Understanding Complex Formations: SPWLA 55th Annual Logging Symposium, Abu Dhabi, United Arab Emirates, May 18-22, 2014, SPWLA-2014-WWW. Dyatlov, G., Kushnir D., and Dashevsky, Yu., 2017. Treatment of singularity in the method of boundary integral equations for 2.5D electromagnetic modeling: Geophysics, Vol. 82(2), pp. E57–E75, DOI: 10.1190/geo2015-0645.1. Eltsov, I., Antonov, Y., Makarov, A., Kashevarov, A., 2011, Invaded Zone Evolution Reconstruction from Logging Data. SEG San Antonio 2011 Annual Meeting, 510 – 513/ Lutfullin, A.A., 2009, Basic Method of Increasing of Recoverable Oil in Russia, Drilling and Life Martinez W.L., Martinez A.R., Solka J.L., 2010, Exploratory Data Analysis with MATLAB, 2nd Edition, CRC Press, Taylor & Francis Group. Miller, P.R., Reid, A.F., and Zuiderwyk, M.A., 1982, QEM*SEM Image Analysis in the Determination Of Modal Assays, Mineral Associations And Mineral Liberation, 14th Int. Min. Proc. Congress, pp VIII-3.1–VII3.2 Molchanov, A.A., Lukyanov, E.E., Rapin, V.A., 2001, Geophysical Surveys in Horizontal Oil Wells, St.Petersburg, MANEB Serra, O., 1985, Sedimentary Environment From Wireline Logs, Schlumberger Limited. Silva, A., Ferraris, P., Barbosa, E., Guedes, A. 2010. Geosteering and/or reservoir characterization the prowess, of newgeneration LWD tools. SPWLA 51st Annual Logging Symposium, Perth. SPWLA-2010-93320. Sviridov, M., Mosin, A., Antonov, Y., Nikitenko, M., Martakov, S., Rabinovich, M., 2014. New Software for Processing of LWD Extradeep Resistivity and Azimuthal Resistivity Data: SPE Reservoir Evaluation & Engineering, Vol. 17(2), pp. 109–127, DOI: 10.2118/160257-PA. Tilsley-Baker, R., Antonov, Y., Martakov, S., Maurer, H-M., Mosin, A, Sviridov, M., Klein, K.; Iversen, M., Barbosa, J., Carneiro, G., 2016. Extradeep-Resistivity Application in Brazil Geosteering Operations Enables Successful Well Landing: SPE Reservoir Evaluation & Engineering, Month: Vol. 19(1), pp. 108–117, DOI: 10.2118/166309-PA. Van der Maaten, L., Hinton, G., 2008, Visualizing Data using t-SNE. Journal of Machine Learning Research 9 (2008) 2579–2605