Journal of Petroleum Science and Engineering 160 (2018) 433441

Contents lists available at ScienceDirect

Journal of Petroleum Science and Engineering

journal homepage: www.elsevier.com/locate/petrol

A method for identifying the thin layer using the wavelet transform of
density logging data
Quanying Zhang a, Feng Zhang a, b, *, Juntao Liu a, Xinguang Wang a, **, Qian Chen a, Liang Zhao a,
Lili Tian a, Yang Wang a
a
School of Geosciences, China University of Petroleum, Qingdao 266580, China
b
Laboratory for Marine Mineral Resources, Qingdao National Laboratory for Marine Science and Technology, Qingdao 266071, China

A R T I C L E I N F O A B S T R A C T

Keywords: In the late stage of oileld development, thin reservoirs become particularly important for oil and gas exploration.
Wavelet transform However, current density logging, as a primary method of reservoir identication, has a lower resolution in
Density logging identifying thin-layers. In this study, a discrete wavelet transform (DWT) is utilized in density logs to identify thin-
Mother wavelet layers. By adopting different Daubechies (dbN) wavelets and decomposition levels, we analyze the approximation
Decomposition level coefcients (cA) and detailed coefcients (cD) and identify the thin-layer signal from detailed coefcients. And
Thin-layer identication
then, we reconstruct a new density curve with enhanced thin-layer signal for identifying the thin layer. Results
show that db4 wavelet and 3 level are the optimum mother wavelet and decomposition level for the density
logging. Detailed coefcients (cD3) from 3rd level decomposition are highly consistent with the thin-layer in-
formation, which is suitable for thin-layer identication. Besides, the reconstructed density curve shows a higher
thin-layer resolution. This method is successfully applied in the oileld, and the thin-layer resolution of density
curve is improved from 30 cm to 15 cm in accordance with microspherically focused logging (RXO).

1. Introduction
On petroleum exploration and exploitation, thin reservoirs have
become important oil and gas sources (Quintal et al., 2008; Zhao et al.,
2013). Density logging, which can identify formation lithology and
divide reservoirs, is crucial to the thin-reservoir identication (Badruz-
zaman, 2014; Zhang et al., 2017). However, the limitation of density
logging resolution makes identifying 20 cm thin-layer become difcult.
Although a series of high-resolution logging tools have been put forward,
the thin-layer resolution of density logging has not been signicantly
improved. Therefore, methods to improve the logging resolution are
fundamentally crucial for the thin-layer identication.
Signal processing technique, as one of the most important methods to
improve the resolution, has been widely used in geophysical exploration
and logging analysis, such as Walsh transform (Lanning and Johnson,
1983; Shi et al., 2006), Fourier analysis (Weedon, 2003), and wavelet
transform (Briqueu et al., 2010; Arabjamaloei et al., 2011; Gaci, 2014).
Maiti and Tiwari (2005) applied the Walsh transform to analyze well log
data with specic criteria and developed an automated method to detect
the lithologic boundaries. Walsh transform is analogous to the Fourier
transform (FT) with a constant window, which is mainly used for
analyzing stationary signals (Pan et al., 2008). While the Fourier analysis
can analyze high-frequency and low-frequency data by using different
windows, but it fails to give spatial localization of the signal (Kadkhodaie
and Rezaee, 2017). By contrast, wavelet transform has spatial frequency
local characteristic and multi-resolution analyzing features. And it can
allow representation of general functions at different scales and positions
in a versatile and sophisticated way.
Wavelet transform not only can identify the formation interfaces, but
also can provide the thickness and spatial localization of different for-
mations, which is of signication for geophysical exploration (Zhao et al.,
2009; Yang et al., 2016; Ge et al., 2014; Karacan and Olea, 2014). Liu
et al. (2006) analyzed on the applicability of the wavelet conversion of
gamma ray curve in the sequence stratigraphic subdivision and obtained
different cycle orders through wavelet analysis of gamma ray (GR) logs.
Pan et al. (2008) analyzed geophysical well logs though wavelet trans-
form and Fourier transform technique and determined formation in-
terfaces, however, their method can only work on the well log signals
which are both uctuation and intensity. Zhang and Song (2010) utilized
the db5 wavelet transform to the GR curves and obtained the time

* Corresponding author. School of Geosciences, China University of Petroleum, Qingdao 266580, China.
** Corresponding author.
E-mail addresses: zhfxy_cn@upc.edu.cn (F. Zhang), wangxg@upc.edu.cn (X. Wang).

https://doi.org/10.1016/j.petrol.2017.10.048
Received 2 May 2017; Received in revised form 13 October 2017; Accepted 18 October 2017
Available online 23 October 2017
0920-4105/ 2017 Elsevier B.V. All rights reserved.
Q. Zhang et al. Journal of Petroleum Science and Engineering 160 (2018) 433441

frequency characteristic for further sequence stratigraphic division with microspherically focused logging.
multiscale analysis. They concluded that wavelet transform can get over
the limitation of manual demarcation by the geologists and present new 2. Methodology of wavelet transform
exploration for the quantitative research of sequence stratigraphy. Zhang
et al. (2011) combined wavelet transforms with conventional log dif- 2.1. Wavelet transform
ferential curves to identify fractured sections. Kadkhodaie and Rezaee
(2017) successfully identied the main sequence boundaries from well Wavelet transform is a signal processing method that converts a signal
log data by using the wavelet transform. Ji et al. (2013) employed gamma to a different form. Such a conversion can observe a signal at different
ray logs and wavelet transform for high resolution sequence stratigraphic scales (from rough to ne), which is very suitable for revealing the hid-
division. Perez-Mu~ noz et al. (2013) identied lithology by the wavelet den information in a signal. The kernel function of wavelet transform is
transform in well logs, but without giving the specic criteria. Ge et al. dened as follow (Pan et al., 2005)
(2015a, 2015b) used improved wavelet transform and exponentially  
weighted moving average to reduce noise of nuclear magnetic resonance 1 tb
a;b t p (1)
(NMR) data and probe the inuential factors. Yang et al. (2016) proposed a a
the application of the synergetic Haar wavelet transform and modied
K-means clustering technique to classify metamorphic rocks using where (t) is called the mother wavelet, a is the scale factor, that de-
CCSD-MH drilling data. termines the wavelength, and b represents the shift of the wavelet
In this study, we applied discrete wavelet transform (DWT) in the (Goupillaud et al., 1984).
density logging to identify thin-layer. Firstly, we analyze the decomposed Continuous wavelet transform (CWT) and discrete wavelet transform
signals (cDs and cAs) from wavelet transform of the density curve by (DWT) are two main types of wavelet transform which is used for the
using different Daubechies (dbN) wavelets and decomposition levels. signal analysis (Chandrasekhar and Rao, 2012; Perez-Mu~ noz et al., 2013;
And the optimum dbN mother wavelet and decomposition level are Yang et al., 2016). A wavelet transform which shifts and dilations are
determined for the wavelet transform of density logs. After that, the thin- continuously varied is called a CWT (Goupillaud et al., 1984). The for-
layer signals are identied by analyzing the scale of decomposed signals. mula is given by
Finally, we reconstruct a high-resolution density curve with enhanced  
thin-layer signals. Besides, the application results of the wavelet trans- 1 tb
Wf a; b p R f t * dt (2)
form in the density logging are presented and compared with the a a

Fig. 1. Block diagram for three-level wavelet transform of a signal.

0.12

db2
0.10 db4
db6
Average error (g/cm 3 )

db8
0.08
db10

0.06

0.04

0.02

0.00

0 1 2 3 4 5 6 7 8 9 10
Decomposition level M
Fig. 2. Relationship between the average error and decomposition level.

434
Q. Zhang et al. Journal of Petroleum Science and Engineering 160 (2018) 433441

The variation of wavelet transformation is represented by an integer decomposed into several sets of approximation wavelet coefcients (cA)
is called a DWT (Daubechies, 1988). In the DWT, the scale and shift from the low-pass lter and detailed wavelet coefcients (cD) from the
parameters are discretized as a 2i and b 2ij, and the formula is high-pass lter. While the wavelet reconstruction is the inverse process of
given by wavelet decomposition, which is the process of reconstructing a signal
t  with the decomposed signals (cA and cD).
1 Taking Fig. 1 as an example (shown in Fig. 1), the Mallat algorithm
Wf i; j p R f t * i  j dt (3)
2i 2 are elaborated. Fig. 1 gives three-level decomposition and reconstruction
of a signal. For convenience, we dene cAM and cDM to represent the
where i, j2Z, f(t) is the origin signal. approximation wavelet coefcients (cA) and detailed wavelet co-
efcients (cD) from the Mth level decomposition or reconstruction,
2.2. Wavelet decomposition and reconstruction respectively. In the rst level, a signal (c0) can be decomposed to a set of
approximation wavelet coefcients (cA1) and a set of detailed wavelet
Mallat algorithm is a mainstream method of decomposing and coefcients (cD1). And in the second level, cA1 is decomposed again into
reconstructing by wavelet transform. And its advantage is the ability of a set of approximation wavelet coefcients (cA2) and a set of detailed
decomposing complex signals into basis signals of nite bandwidth, and wavelet coefcients (cD2). Similarly, cA2 also can produce cA3 and cD3
then reconstructing them again with very little loss of information in the third level. Finally, a signal (c0) can be decomposed to three sets of
(Mallat, 1989). In the wavelet decomposition, a signal can be detailed wavelet coefcients (cD1, cD2 and cD3) and a set of

Fig. 3. Comparisons of the low-resolution signals and the original density curve.

435
Q. Zhang et al.
approximation wavelet coefcients (cA3) by the three-level wavelet
transform. Note that, for the Mallat algorithmthe, the object of the next
level decomposition is always the approximation wavelet coefcients
(cA) from the previous level decomposition, while the detailed wavelet
coefcients (cD) from the previous level decomposition will not be
affected by the subsequent decompositions.
We apply this process to study the spatial information content of the
density log and seek to improve the vertical resolution if possible. In the
process, the density log can be decomposed to approximation coefcients
and detailed coefcients. And the detailed coefcients, including the
information from thin layers, can be used for reecting the thin layers,
while the approximation coefcients correspond to the surround-
ing rocks.
3. Application of wavelet transform in the density logging
In the density logging, the thin-layer information is always covered by
the information of surrounding rock, which is account for the low reso-
lution. By the wavelet transform, the thin-layer information can be
revealed to be used for improving the thin-layer resolution. While the
application results of wavelet transform are determined by the mother
wavelet, decomposition level and identication of thin-layer signals.
3.1. Determining the optimum mother wavelet and decomposition level
In geophysical exploration, Daubechies (dbN) wavelet and Haar
wavelet are the most common orthogonal wavelet bases used for the
wavelet transform (Pan et al., 2008; Zhang and Song, 2010; Hon orio
et al., 2012; Yang et al., 2016; Taherdangkoo and Abdideh, 2016;
Mukherjee et al., 2016). DbN wavelet is a set of wavelets developed by
Daubechies where the wavelets are denoted by dbN where N is the serial
number of the wavelet function (N 1, 2 , 10) (Daubechies, 1988).
Particularly when serial number N is equal to 1, the db1 is Haar wavelet.
Therefore, we choose Daubechies (dbN) wavelets to make wavelet
transform. To employ this process to density log, the most applicable of
the dbN wavelets must be determined.
0.030
0.028
A verage error (g/cm 3 )
0.026
0.024
0.022
0.020
0 1
Fig. 4. Relationship between the average error and wavelet serial number N.
Journal of Petroleum Science and Engineering 160 (2018) 433441
An average error function (Er) is dened as a selection criterion for
the dbN wavelet and decomposition level M,
X
k
Er jcAMi  c0i j=k (4)
i
where k is the number of the logging data points, i is the ith data point,
cAM is the approximation wavelet coefcients (cA) from the Mth level
decomposition. The average error (Er) can reect the difference between
the origin signal and decomposed signal.
3.1.1. Decomposition level M
Wavelet transforms of density log are carried out using db2, db4, db6
and db8 wavelets, respectively, and low-resolution signal curves (cA)
from different decomposition levels are derived.
Fig. 2 shows the relationship between the average error and decom-
position level. As the decomposition level increases, the average error
becomes larger. That is because more detailed information is separated
from the original density curve. Besides, the relationships between the
average error and decomposition level using different dbN wavelets are
almost the same. That means the wavelet transforms using different dbN
wavelets have little impact on the determination of optimum decompo-
sition level. Therefore, we only need take a single dbN wavelet as an
example to optimize decomposition level, instead of all dbN wavelets.
Taking the db4 wavelet as an example, the low-resolution curves (cA)
from different decomposition levels are compared with the original
density curve (c0), shown in Fig. 3.
Fig. 3 shows that the difference between the low-resolution curve
(cA) and the original density curve becomes larger and larger, as the
decomposition level increases. For the rst and second level decompo-
sition, the difference is relatively small and the low-resolution curve (cA1
or cA2) almost coincides with the original curve, which means the thin-
layer signal is not successfully extracted into the high-resolution signals
(cD1 and cD2). While for the 3rd level decomposition, there is obvious
thin-layer information extracted to the high-resolution signals (cD3). As
for the fourth or higher level decomposition, there is some information of
2 3 4 5 6 7 8 9 10 11
Serial number N of dbN wavelet
436
Q. Zhang et al.
the surrounding rock being extracted from the low-resolution signals into
the high-resolution signals.
Theoretically, a signal can be decomposed innitely as long as
decomposition level M is large enough. In fact, when the desired signals
are extracted from the original signal to the high-resolution signals (cD),
the decomposition level M is sufcient for data processing. For the
density log, the thin-layer signal have been extracted from cA2 to the
high-resolution signals (cD3) when the decomposition level M reaches 3.
For four or higher level decomposition, only the low-resolution curve
(cA3) from 3 level decomposition is decomposed again, which is not
helpful for data processing. Therefore, we choose the three level as the
optimum decomposition level.
3.1.2. Mother wavelet
The correlation between original signal and wavelet function is the
crucial standard for mother wavelet selection (Zhang et al., 2011; Yang
et al., 2016), which also can be reected by the correlation between
wavelet coefcients and original signals.
In the paper, we choose the average error of decomposed signal (cA3)
and original signal as the criterion for mother wavelet selection. The
Fig. 5. Comparisons of the low-resolution signals and the original density curve.
437
Journal of Petroleum Science and Engineering 160 (2018) 433441
smaller the error, the higher the correlation. Decomposition level M is set
to 3, the wavelet transforms are carried out by using db1 to db10 wavelet
functions, respectively. Fig. 4 shows the relationship between the
average error and the serial number N of dbN wavelet. And the low-
decomposed signal (cA3) from different dbN wavelets is compared
with the original density curve (c0), shown in Fig. 5.
As shown in Figs. 4 and 5, with the increasing serial number, the
decomposed signal (cA3) becomes smoother and the correlation becomes
higher. For the db1 to db3 wavelet functions, corresponding decomposed
signals (cA3) are broken lines, and the correlation is relatively lower.
Especially for db1 wavelet, the decomposed signal is the square-wave
signal. When the serial number N reaches 4 or higher, the decomposed
signal becomes a smooth curve and have a higher correlation with
original signals. Besides, the average error reaches the minimum when
the serial number N is 4, and it hardly changes when serial number N
continues to increase. That means the decomposed signals from the db4
to db10 wavelets all have a higher correlation with the density log in
the waveform.
However, for the dbN wavelet, the larger the serial number N, the
weaker the local characteristic of the wavelet function (Liu et al., 2006).
Q. Zhang et al.
To meet the need for identifying the thin-layer, the sufcient local
characteristics of wavelet function must be ensured. Therefore, the db4
wavelet is chosen as the optimum mother wavelet, which has enough
correlation and local characteristics.
3.2. Identication of thin-layer information
According to analyses of the decomposed levels and mother wavelets,
the three-level wavelet transform of density log is carried out by db4
wavelet. And the density log is decomposed into three high-resolution
signal curves (cD1, cD2 and cD3) and a low-resolution signal curve
(cA3), shown in Fig. 6.
As shown in Fig. 6, the density range of the low-resolution curve
(cA3) is 2.02.7 g/cm3, and it has the same magnitude order as the
original curve. That is because the low-resolution curve mainly reects
the density of surrounding rocks. For the high-resolution curve (cD1)
from the rst level decomposition, the density range and wave scale is
0.005 to 0.005 g/cm3 and 020 cm, respectively, and the corre-
sponding resolution (half of the wave scale) is less than 010 cm. So the
Fig. 6. Decomposition signal curves from the three-level wavelet transform.
438
Journal of Petroleum Science and Engineering 160 (2018) 433441
cD1 mainly includes systematic errors and random errors, which can be
used for the curve ltering rather than thin-layer identication. Simi-
larly, the density range and wave scale of the high-resolution curve (cD2)
from the second level decomposition is 0.01 to 0.01 g/cm3 and
2030 cm, respectively, and the corresponding resolution is
10 cm15 cm. It may include the systematic error and interference in-
formation from the wellbore and ultra-thin layers, which can not be used
to identify thin layers. As for the high-resolution signal curve (cD3) from
the third level decomposition, the density range and wave scale is 0.04
to 0.04 g/cm3 and 3040 cm, respectively, and the corresponding reso-
lution is 15 cm20 cm. The resolution of cD3 curve is highly consistent
with the thickness of thin layers. Therefore, the decomposition curve cD3
can be used for thin-layer identication.
Fig. 7 gives the power spectrums of the cD1, cD 2 and cD 3 curves by a
Fourier transform. It can be seen that three curves have an obvious dif-
ference in the signal power distribution and magnitude order. Compared
to cD2 and cD3 curves, the signal power of cD1 is mainly distributed in
the higher frequency range (more than 1.25 Hz), while the signal power
of cD2 is mainly distributed around 0.7 Hz and the cD3 is 0.33 Hz.
Q. Zhang et al.
-30
-40
Power spectrum (dB/Hz)
-50
-60
-70
-80
-90
-100
0.00 0.25
Fig. 7. Power spectrums of the cD1, cD 2 and cD 3 curves (Sampling frequency 5 Hz, sampling points 5272).
Besides, the signal power of cD3 is 10 dB/Hz greater than that of cD2, and
20 dB/Hz greater than that of cD1. Besides, it can be seen that the signal
power of thin-layer is an absolute advantage in the cD3 curve.
3.3. Reconstruction of the high-resolution density curve
According to Mallat wavelet transform, the relationship of original
density curve (c0) and decomposed signals (cA and cD) can be given
as follows,
c0 cA1 cD1 cA2 cD2 cD1 cA3 cD3 cD2 cD1
where cD3 curve contains the thin-layer information. In order to in-
crease the proportion of thin-layer information in the density curve, we
make energy-compensation for cD3 curve by adjusting the weight co-
efcients, shown as Eq. (6),
c1 cA1 w1 cD1 cA2 w2 cD2 w1 cD1
cA3 w3 cD3 w2 cD2 w1 cD1
where c1 is the reconstructed density curve, w1, w2 and w3 are the weight
coefcients of cD1, cD2 and cD3, respectively. In the reconstruction
process, weight coefcients of cD1, cD2 and cD3 are set to 1.0, 1.0 and
2.5, respectively. The choice of weight coefcients not only can enhance
the thin layer signal as much as possible, but also have no larger impact
on the density of surrounding rock.
4. Results and discussion
Fig. 8 displays a eld case of the wavelet transform application in the
density logging. In the case, all the logging curves are obtained from an
open-hole well, including natural gamma ray (GR), high-resolution
borehole diameter (HCAL), acoustic (AC), high-resolution acoustic
(HAC), deep laterolog (LLD), shallow laterolog (LLS), ushed zone for-
mation resistivity (RXO) and density (DEN) curves. Among them, RXO
curve has 15 cm vertical resolution, which is mainly used for thin layer
Journal of Petroleum Science and Engineering 160 (2018) 433441
cD1
cD2
cD3
0.50 0.75 1.00 1.25 1.50 1.75 2.00 2.25 2.50
Frequence (Hz)
division, while DEN curve has only 30 cm vertical resolution. By the
wavelet transform, thin-layer information is extracted from DEN curve
and enhanced by adjusting weight coefcients. Finally, a new density
curve (DENS) is reconstructed with the enhanced thin-layer signal.
As shown in Fig. 8, compared with the original density curve (DEN),
the reconstructed density curve (in track 6 in the red dash) have a sig-
nicant enhancement in the resolution, particularly in the identifying
thin layers of 15 cm-20 cm thickness. For example, according to the
analysis of RXO, there are four thin layers existing in the range of
3003 m-3005 m. However, the original density curve (in track 6 in the
(5) black line) only shows two layers. After wavelet transform processing,
the reconstructed density curve (DENS) shows four thin layers. Besides,
in the range of 3006.5m3007.1 m and 30343035, RXO curve shows
two thin layers which are not displayed on the original density curve, but
the reconstructed density curve shows the corresponding thin layers. The
results of thin-layer analysis show that the resolution of density curve is
improved from 30 cm to 15 cm by the wavelet transform.
(6) 5. Conclusions
In this paper, the discrete wavelet transform is applied to identify the
thin-layer using density logging data. By adopting different Daubechies
(dbN) wavelets and decomposition levels, we analyze the approximation
coefcients (cA) and detailed coefcients (cD). By the average error and
correlation between the decomposed signals and density log curve, the
db4 wavelet and 3 level decomposition are determined as the optimum
mother wavelet and decomposition level for the density logging. And
then, the decomposed signals (cA3) are considered to contain thin-layer
information by analyzing the vertical resolution. Finally, we reconstruct
a high-resolution density curve to identify thin layers.
The process is successfully applied in the oileld. The results show
that the reconstructed density curve shows a signicant enhancement in
identifying the thin layers compared with the original density curve, and
the thin-layer resolution of density curve is improved from 30 cm
to 15 cm.
439
Q. Zhang et al.
Fig. 8. Field example.
Acknowledgments
This work was supported by National Natural Science Foundation of
China (41574119, and 41504099); the National Major Oil and Gas
Special Fund of China (2017ZX05019005-004); the Natural Science
Foundation of Shandong Province (ZR2015DQ003); and the Special
Fund of Central College Basic Research Operating Expenses
(15CX06008A, 17CX02070).
440
Journal of Petroleum Science and Engineering 160 (2018) 433441
References
Arabjamaloei, R., Edalatkha, S., Jamshidi, E., et al., 2011. Exact lithologic boundary
detection based on wavelet transform analysis and real-time investigation of facies
discontinuities using drilling data. Pet. Sci. Technol. 29, 569578.
Badruzzaman, A., 2014. An assessment of fundamentals of nuclear-based alternatives to
conventional chemical-source bulk density measurement. Petrophysics 55 (5),
415434.
Briqueu, L., Zaourar, N., Lauer-Leredde, C., Hamoudi, M., 2010. Wavelet based multiscale
analysis of geophysical downhole measurements: application to a clayey siliclastic
sequence. J. Pet. Sci. Eng. 71 (34), 112120.
Q. Zhang et al.
Chandrasekhar, E., Rao, V.E., 2012. Wavelet analysis of geophysical well-log data of
Bombay offshore basin, India. Math. Geosci. 44 (8), 901928.
Daubechies, I., 1988. Orthonormal bases of compactly supported wavelets. Commun.
Pure Appl. Math. 41 (7), 909996.
Ge, X., Fan, Y., Zhu, X., et al., 2014. A method to differentiate degree of volcanic reservoir
fracture development using conventional well logging dataan application of kernel
principal component analysis (KPCA) and multifractal detrended uctuation analysis
(MFDFA). Sel. Top. Appl. Earth Obs. Remote Sens. IEEE J. 7 (12), 49724978.
Ge, X., Fan, Y., Li, J., et al., 2015a. Noise reduction of nuclear magnetic resonance (NMR)
transversal data using improved wavelet transform and exponentially weighted
moving average (EWMA). J. Magn. Reson. 251, 7183.
Ge, X., Fan, Y., Chen, H., et al., 2015b. Probing the inuential factors of NMR T 1T 2
spectra in the characterization of the kerogen by numerical simulation. J. Magn.
Reson. 260, 5466.
Gaci, S., 2014. The use of wavelet-based denoising techniques to enhance the rst-arrival
picking on seismic traces. IEEE Trans. Geosci. Remote Sens. 52 (8), 45584563.
Goupillaud, R.A., Grossmann, A., Morlet, J., 1984. Cycle-octave and related transform in
seismic signal analysis. Geoexploration 23 (1), 85102.
Hon orio, B.C.Z., Drummond, R.D., Vidal, A.C., et al., 2012. Well log denoising and
geological enhancement based on discrete wavelet transform and hybrid
thresholding. Energy Explor. Exploit. 30 (3), 417433.
Ji, D.W., Li, Ji, Lu, G.D., 2013. Application of wavelet transform in high-resolution
sequence stratigraphic division. Adv. Mater. Res. 772, 823827.
Kadkhodaie, A., Rezaee, R., 2017. Intelligent sequence stratigraphy through a wavelet-
based decomposition of well log data. J. Natural Gas Sci. Eng. 40, 3850.
Olea, R.A., 2014. Inference of strata separation and gas emission paths in
Karacan, C.O.,
longwall overburden using continuous wavelet transform of well logs and
geostatistical simulation. J. Appl. Geophys. 105, 147158.
Lanning, E.N., Jonson, D.J., 1983. Automatic identication of rock boundaries; an
application of the Walsh transform to geophysical well-log analysis. Geophysics 48,
197205.
Liu, J., Han, S., Ma, J., et al., 2006. Application of wavelet analysis in seismic data
denoising. Prog. Geophys. 21 (2), 541545.
Maiti, Tiwari, R.K., 2005. Automatic detection of lithologic boundaries using the Walsh
transform. A case study from the KTB borehole. Comput. Geosci 31, 948955.
441
Journal of Petroleum Science and Engineering 160 (2018) 433441
Mallat, S.G., 1989. Multi resolution approximation and wavelets. Trans. Amer. Math. Soc.
315, 6988.
Mukherjee, B., Srivardhan, V., Roy, P.N.S., 2016. Identication of formation interfaces by
using wavelet and Fourier transforms. J. Appl. Geophys. 128, 140149.
Pan, Q., Zhang, L., Meng, J.L., 2005. The Wavelet Analysis Method and its Application.
Tsinghua University Press, Beijing, China.
Pan, S.Y., Hsieh, B.Z., Lu, M.T., et al., 2008. Identication of stratigraphic formation
interfaces using wavelet and Fourier transforms. Comput. Geosci. 34, 7792.
Perez-Mu~ noz, T., Velasco-Hernandez, J., Hernandez-Martinez, E., 2013. Wavelet
transform analysis for lithological characteristics identication in siliciclastic oil
elds. J. Appl. Geophys. 98, 298308.
Quintal, B., Schmalholz, S.M., Podladchikov, Y.Y., 2008. Low-frequency reections from a
thin layer with high attenuation caused by interlayer ow. Geophysics 74 (1), 1523.
Shi, Q.J., Wang, Y.J., Sun, Z.Y., et al., 2006. Joint application of wavelet transform and
Walsh transform for automatic segmentation of well log. Zhongguo Shi You Daxue
Xuebao J. China Univ. Petrol. Ed. Nat. Sci. 30 (2), 138142.
Taherdangkoo, R., Abdideh, M., 2016. Application of wavelet transform to detect
fractured zones using conventional well logs data (case study: southwest of Iran). Int.
J. Petrol. Eng. 2 (2), 125139.
Weedon, G.P., 2003. Time-Series Analysis and Cyclostratigraphy. Cambridge University
Press, Aberdeen, UK.
Yang, H., Pan, H., Ma, H., et al., 2016. Performance of the synergetic wavelet transform
and modied K-means clustering in lithology classication using nuclear log. J. Pet.
Sci. Eng. 144, 19.
Zhao, D.W., Wang, J., Gates, I.D., 2013. Optimized solvent-aided steam-ooding strategy
for recovery of thin heavy oil reservoirs. Fuel 112, 5059.
Zhao, W., Qiu, L., Jiang, Z.X., et al., 2009. Application of wavelet analysis in high-
resolution sequence unit division. J. China Univ. Petrol. Ed. Nat. Sci. 2, 6.
Zhang, J., Song, A., 2010. Application of wavelet analysis in sequence stratigraphic
division of glutenite sediments. In: Challenges in Environmental Science and
Computer Engineering (CESCE), 2010 International Conference on. Vol. 2. IEEE.
Zhang, Q., Zhang, F., Liu, J., et al., 2017. A method of determining formation density
based on fast-neutron gamma coupled eld theory. Petrophysics 58 (4), 411425.
Zhang, X.F., Pan, B.Z., Wang, F., et al., 2011. A study of wavelet transforms applied for
fracture identication and fracture density evaluation. Appl. Geophys. 8 (2),
164169.