Application of Seismic Multi-Attribute Inversion Based on

Sensor Network and Deep Neural Networks

YuanyuanWang1,2*, CuiWang1

1 Institute of Resource and Environment, Henan Polytechnic University, Jiaozuo 454003, China

2 Henan Key Laboratory of Biogenic Traces&Sedimentary Minerals of Henan Province, Jiaozuo

454003, China

Email:[email protected]

Abstract:Seismic attributes, which are extracted from seismic information, are physical indexes used
specificallyfor the measurement of geometric, dynamic, or statistical characteristics of seismic data.

Current methods for seismic multi-attribute inversion include linear and nonlinear methods. By

adopting the wireless module of NFC24l01, combined with the seismic data acquisition sensor,

constitutes an intelligent network sensor, and then it sends the collected data to the topmost machine

for analysis.Methods for the nonlinear inversion of seismic multi-attributes usually employ tools such

as neural networks and support vector machines (SVMs)for mapping. Hence, inversion results obtained

via nonlinear methodsare more accurate than those obtained via linear methods. In this work, with

spontaneous-potential (SP) curves as the objective of nonlinear inversion, an optimized seismic

attribute combination for the inversion of SP curves was identified, and the nonlinear inversion of

seismic multi-attributes was achieved via the use of a deep neural network (DNN) to obtain 3D SP data.

Finally, the foresetting process of a sand body of the intermediate section in Member 3 of the Shahejie

Formation in the Dongying Delta was illustrated via the horizon slice of the SP data.

Keywords: Sensor network; seismic multi-attribute inversion; deep neural network; delta foreset

1. Introduction

Seismic attributes are closely related to reservoir parameters. However, the seismic data acquisition

way is less concerned. Wireless sensor network (WSN) is a set of advanced technology in the

integration of information technology, it is not only a universal computer, and also the effective

1
extension of human senses, the data transmission and exchange as the center, as a kind of

self-organized and distributed network, a large number of sensor nodes in sensor network to provide

the data, even in very poor conditions, also can do real-time monitoring, cooperative awareness and

information collection, and send the monitoring data.This makes wireless sensor networks widely used

in many fields.

Seismic attribute analysis involves the prediction of reservoir parametersbased on seismic

attributes[1-3] and has played an increasingly important role in the exploration and exploitation of oil

and gas. Seismic attributes are closely related to reservoir parameters[4], but they have no direct

analytic relation. The basic concept of the nonlinear inversion of seismic multi-attributes[5,6]is to

establish a nonlinear relation of seismic multi-attributes and reservoir parameters via mathematical

methods(e.g., multi-variable stepwise regression, neural networks, support vector machines (SVMs),

and the co-Kriging method[7-9] and obtain information required by reservoir prediction.

In this work, spontaneous-potential (SP) curves were employed as the objective of the nonlinear

inversion of seismic multi-attributes.First, an optimized seismic attribute combination for the inversion

of SP curves was identified via nonlinear regression. Then, 3D SP data was obtained by the nonlinear

inversion of seismic multi-attributes based on a deep neural network (DNN). Finally, the foresetting

process of a sand body of the intermediate section in Member 3 of the Shahejie Formation in the

Dongying Delta was illustrated by the horizon slice of the SP data body. The practical application

results indicate the good application performance and great potential of the proposed nonlinear

inversion of seismic multi-attributes.

2. Principles and methods

2.1 Optimization of seismic attributes

The optimization of seismic attributes aims to retrieve the optimized attribute combination from

various seismic attributes via stepwise regression methods[10]. First, the correlations between each

single attribute and the conversion curve were calculated, and the attribute with the highest correlation

was retained. Then, the second, third, …, xth (where x refers to the designated maximum attribute

number) optimized attributes were retrieved from the remaining attributes according to the root mean

square criterion.
2
2.2 Convolution operator

The value of the parameter to be inverted at the sampling point of the curve is dependent on both

its attribute and several points above and below the corresponding seismic attribute point. Therefore,

the value of the parameter to be inverted at the sampling point of the curve is correlated to points above

and below the corresponding seismic attribute point to further improve the stability of the

multi-attribute inversion solution. This can be achieved by the convolution operator:

L = 0 + w1m  A1 + w2m  A2 + ... + wnm  An , (1)

m m
where L refers to the parameter to be inverted, ω0is a constant, w1 , w2 , , wnm are convolution

operators, m refers to the length of the convolution operator, and A1 , A2 , , An are N attributes

retrieved via the optimization of attribute combination.

Eq. (1) can be simplified as:

L = Aw . (2)

For accurate data model parameters, the convolution operator can be expressed as:

w = A−1L . (3)

However, practical data cannot accurately fit the model due to experimental error:

L = Aw + ei , (4)

where ei refers to the error.

The most effective way to obtain the unique solution of a model parameter is the minimization of

the sum of square of errors:

5 12
q = eT e =  ( Li −  Aij w j )2 (5)
i =1 j =1

or

q = ( L − Aw)T ( L − Aw) , (6)

3
where q refers to the sum of square of errors, T refers to the matrix transposition, Li refers to the

parameter to be inverted, Aij refers to the seismic attribute, and wj refers to the convolution operator.

The derivation of Eq. (6) leads to:

q   L L − L Aw − w A L + w A Aw
T T T T T T

= . (7)
w j w j

The canonical equation is:

2 AT Aw = 2 AT L . (8)

The estimated least squares solution of the parameter can be obtained:

−1
wˆ =  AT A AT L . (9)

In this way, the convolution operator is obtained.

2.3 Deep neural network

A DNN is essentially a modified version of conventional neural networks,and its structure is

illustrated in Fig. 1. A typical DNN consists of an input layer, output layer, and several hidden layers.

Each layer has several nodes that are connected by weights, and the results of the output layer are

determined by the weights[11,12]. In this study, a DNN consisting of one input layer, one output layer,

and three hidden layers was used. Herein, a convolution operator is involved, and the seismic attribute

is convoluted with the convolution operator before being used as the input of the DNN.

x1
x2 y1
x3 y2
…

xn ym

Input layer Hidden layer Output layer

4
 Fig. 1. Schematic of the DNN structure.

The optimized weights of nodes in hidden layers can be calculated via the learning process of the

DNN, and network training can be realized by employing a training set as the input of the DNN.

Weight determination can be regarded as a nonlinear optimization process, the objective of which is the

minimization of the mean square error of the practical target log value and predicted target log value.

Solutions can be obtained by methods such as back propagation[11,12].

In summary, seismic multi-attribute inversion involves the establishment of a nonlinear relation of

seismic multi-attributes and SP via mathematical methods. The SPs of strata are calculated based on the

log data of known wells, and the lithology of strata can be identified by the inversion of SP parameters

based on seismic data[13,14].

3. Practical applications

Delta is a product of the combination of water basin, the river input, tides, waves, and the impact

of factors such as tectonic and climate, with complex composition and sedimentary characteristics,

sedimentary environment. Delta is a kind of very important sedimentary environment. Many oilfields

in China, such as daqing oil field, shengli oilfield, changqing gas field, xinjiang oilfield, delta sand

body are main reservoir and is an important part in oil and gas gathering. For Member 3 of the Shahejie

Formation in the Dongying Delta, the sedimentary period was the peak of fault subsidence[15,16],

during which large-scale delta systems were developed. In previous studies[17,18], the intermediate

section in Member 3 of the Shahejie Formation in Dongying Delta was divided into nine stages

(labeled as z1 through z9). However, the distribution range and location of the convolutional sand body

of the intermediate section in Member 3 of the Shahejie Formation in the Dongying Delta remain

unclear. In this study, the foresetting range and process of the sand body of the intermediate section in

Member 3 of the Shahejie Formation in the Dongying Delta were illustrated by the horizon slice of the

SP data obtained via seismic multi-attribute inversion.

According to the velocity curves of the intermediate section in Member 3 of the Shahejie

Formation in the Dongying Delta forest, the velocity of sandstone is slightly higher than that of

mudstone, while characteristic impedances of sandstone and mudstone are not significantly different.

Indeed, the longitudinal wave velocities of sandstone and mudstone are concentrated in the range of
5
3000-4500 m/s, as shown in Fig. 2. Therefore, it is highly difficult to directly distinguish sandstone and

mudstone via wave velocity or characteristic impedance. Further analysis of the SP curves reveals that

the SPs of mudstone are concentrated in a high-value zone, while those of sandstone are concentrated

in a low-value zone. Therefore, it can be concluded that sandstone and mudstone can be effectively

distinguished, and the forest sand body in the delta can be identified by SPs.

Sandstone
5000
Mudstone

4500
Wave velocity(m/s)

4000

3500

3000
(m/s )

2500

2000
-10 0 10 20 30 40 50 60 70

Spontaneous-potential(mv)

Fig. 2. Wave velocity vs.SP of sandstone and mudstone in the target area.

In the target area, inversion wells were selected to be uniformly distributed. To enhance the

inversion performance, the SP curves of some logs involved in multi-attribute inversion were

normalized, as presented in Fig. 3.

6
 1400 1600 1600 2000

1600 1800 2200

1800
1800 2400
2000

2000 2000 2600

2200
Depth(m)

2200 2800
2200
2400
(m)

2400 3000
2600 2400
2600 3200

2800
2800 2600 3400

3000 3000 3600

2800
3200 3200 3800

3400 3400 3000 4000

0 0.5 1 0 0.5 1 0 0.5 1 0 0.5 1
SP SP SP SP
(a)Niu103 (b) Niu26 (c) Wang580 (d) Wang78

Fig. 3. SP curves of some logs involved in multi-attribute inversion (after normalization).

The optimization of the seismic attribute combination was first conducted. In Fig. 4, the x- and y-

axes denote the quantity of seismic attributes and the error, respectively. As the quantity of seismic

attributes increases, the training error decreases, while the testing error decreases and then

increases,indicating that abundant seismic attributes lead to degraded system stability and reduced

prediction accuracy. Therefore, the first five seismic attributes were employed as the input of the DNN,

as exhibited in Table 1.

Training error
0.24
Testing error

0.22
Error(mv)

0.20

0.18
1 2 3 4 5 6 7 8
Quantity of seismic attributes

Fig. 4. Quantity of seismic attributes vs. error.

7
 Table 1. The first five seismic attributes used for multi-attribute inversion.

Seismic attribute Training error Cross-test error

1 Average frequency 0.209 0.221

2 Absolute amplitude 0.204 0.226

3 Transient phase 0.201 0.227

4 Band-pass filtering 0.200 0.223

5 Apparent polarity 0.198 0.220

Fig. 5 shows an east-west well section of the intermediate section in Member 3 of the Shahejie

Formation in the Dongying Delta as obtained by multi-attribute SP inversion. As is evident, the

inversion results are highly consistent with the SP curves of all wells. Additionally, section-by-section

inversion reveals that this delta is a prograded delta with provenance in the southeast area. The

development of the subfacies are delta plain, delta front, and front delta, successively.

CDP
3100 3500 3900 4280
1200

Niu 26

Niu 80
Wang78

Delta plain
Times(ms)

2000

Delta front

Front delta

2800

Fig. 5. Well section of the intermediate section in Member 3 of the Shahejie Formation in the Dongying

Delta as obtained by multi-attribute SP inversion.

Fig. 6 illustrates the slice of the intermediate section in Member 3 of the Shahejie Formation in the

Dongying Delta as obtained by multi-attribute SP inversion. As can be observed, the foresting process

of the delta sand body consists of several stages. At the z7 stage, the target area was dominated by a

deep/semi-deep lake, delta lobes coming from the southeast and northeast areas pronograde towards the

8
northwest and borders, and foreset sand body was concentrated in the east area. At the z5 stage, the

target area was still dominated by a deep/semi-deep lake, while the delta sand body significantly

pronograde towards the northwest. Additionally, provenance from the Weibei Hump delta in the

southeast provided sufficient supply; progradation was significant, and the prograded sand body

crossed the middle part of the target area. At the z3stage, the lake basin area shrank drastically, and

deep/semi-deep lake sediments were only distributed in the northwest part of the target area.

Additionally, delta lobes from the northeast and south are asmerged and prograded westwards,

provenance from the Weibei Hump provided continuous supply via two main river channels,and the

sand body thickness was maximized in this area. At the z1 stage, the foreset sand body prograded to the

west part of the target area, deep/semi-deep lake sediment in the target area nearly disappeared, delta

lobes from northeast and south areas merged,and the pattern was not very sharp. Additionally, river

channels for provenance supply intertwined with each other.

9
 Horizontal line

0.9
The
z7

0.7

Spontaneous-potential
Foreset
Vertical
line

z5

0.5

Sand

z3

0.3

Body

z1

Fig. 6. Horizon slice of the intermediate section in Member 3 of the Shahejie Formation in the

10
Dongying Delta as obtained by multi-attribute SP inversion. The yellow is sand; green is silt; purple is

mud and blue is silty-mud.

4. Conclusions

Seismic multi-attribute inversion has been an effective method for lithological inversion,

especially for areas where it is difficult to distinguish sandstone and mudstone bycharacteristic

impedances. In this study, the foresetting process of the sand body of the intermediate section in

Member 3 of the Shahejie Formation in the Dongying Delta was described by the nonlinear inversion

of seismic multi-attributes with SP curves as the objective of nonlinear inversion. Seismic

multi-attribute nonlinear inversion is essentially a modified version of conventional seismic inversion

methods. This method does not rely on any specific forward model or seismic wavelets, and is suitable

for many applications.

List of abbreviations
(SVMs) Support Vector Machine Svm

Declarations

Availability of data and material

The datasets used and/or analyzed during the current study are available from the corresponding

author on reasonable request.

Competing interests
declares that he has no conflict of interest.

* Research involving human participants and/or animals

Ethical approval: This article does not contain any studies with human participants or
animals performed by any of the authors.

* Informed consent

11
All authors agree to submit this version and claim that no part of this manuscript has
been published or submitted elsewhere.

Funding
This paper was supported by the National Science Foundation of China (Nos., 41602114,
41602112, 41772104) and the Henan Key Laboratory of Biogenic Traces.

Authors' contributions
YuanyuanWang was responsible for the collection of experimental data, and CuiWang was responsible
for the simulation and simulation of the experiment

Acknowledgements
This paper was supported by the National Science Foundation of China (Nos., 41602114,
41602112, 41772104) and the Henan Key Laboratory of Biogenic Traces. Thanks to
PeijieYang’s valuable advice who from Petroleum Exploration & Development Research
Institute of Shengli Oil Field branch company, China Petroleum & Chemical Corporation,
Dongying, Shandong, China.

Availability of data and material

The datasets used and analysed during the current study are available from the corresponding
author.

References

1. Chen Q, Sideney S.,“Seismic attribute technology for reservoir forecasting and monitoring,”The

Leading Edge, vol.16,no.5,pp.445-456, 1997.

2. BarnesA. Seismic attributes in your facies. CSEG Recorder, vol.9,pp.41-47, 2001.

3. Lawrence P, Aramco S, Dhahran, et al.,“Seismic attributes in the characterization of small-scale

reservoir in Abqaiq Field,”The Leading Edge, vol. 17,no. 4,pp. 521-525, 1998.

4. Liu Z, and Liu J.,“Seismic-controlled nonlinear extrapolation of well parameters using neural

networks,”Geophysics, vol. 63,no. 6,pp. 2035-2041, 1998.

5. Liu S, Xie Z, Liu J, et al.,“Application of pre-stack multi-attribute inversion technology on

igneous rock investigation in Nanpu oilfield,”Geophysical Prospecting for Petroleum, vol. 47,no.

12
 1,pp. 83-93, 2008.

6. Ming J, Huang K, Zhang J, et al.,“An application of neural-network seismic inversion with

multiple attributes to horizontal drilling in NB oilfield,”China Offshore Oil and Gas, vol. 18,

no.6,pp. 382-384, 2006.

7. Le Y, Wang Y.,“Nonparametric regression applied in porosity parameter prediction,”Chinese

Journal of Geology, vol. 37,no. 1,pp. 118-125, 2002.

8. Hampson D, Yu B, Liu X.,“Use of multiattribute transforms to predict log properties from seismic

data,”Progress in Exploration Geophysics, vol. 25,no. 3,pp. 65-75, 2002.

9. Yang P, Yin X.,“Pre-stack seismic inversion method based on support vector machine,”Journal of

China University of Petroleum (Edition of Natural Science), vol. 32,no. 1,pp. 37-41, 2008.

10. Kong G.,“Extraction and optimization of pre-stack seismic attribute,”China University of

Petroleum (Huadong), 2008.

11. Jiao L.,“Theories of neural network systems,”Xidian University Press, 1990.

12. Kirkpatrick S, Gelatt C, Vecchi M.,“Optimization by simulated annealing,”Science, vol. 220 no.

4598,pp. 671-680, 1983.

13. Qiu G, Wang J, Zhang X, et al.,“Preliminary study on stratigraphic architecture of

middle-shasanDongying Delta and its significance to Hydrocarbon Exploration Dongying

Delta,”Acta SedimentologicaSinica, vol. 19,no. 4,pp. 569-575, 2001.

14. Liu H, Yang P, Liu S, et al.,“Seismic Wheeler transform and its application,”Geophysical

Prospecting for Petroleum, vol. 51,no. 1,pp. 51-55, 2012.

15. Mu X, Lu X, Yang P.,“High resolution time-frequency analysis method based on high-order

statistics,”Petroleum Geology and Recovery Efficiency, vol. 16, no. 6,pp. 56-59, 2009.

16. Liang H., Zou J. Zuo K., Khan Muhammad J.,“An improved genetic algorithm optimization fuzzy

controller applied to the wellhead back pressure control system,” Mechanical Systems and Signal

Processing, vol. 142, pp. 106708, 2020.

13
17. Chonghuan Xu, “A novel recommendation method based on social network using matrix

factorization technique,” Information processing & management,vol. 54, no. 3, pp. 463-474, 2018.

18. Liang H., Zou D., Li Z., Khan Muhammad J., Lu Y. ,“Dynamic evaluation of drilling leakage risk

based on fuzzy theory and PSO-SVR algorithm,” Future Generation Computer Systems, vol. 95,

pp. 454-466, 2019.

Figure 1:

Title: Schematic of the DNN structure

Legend: Figure 1 show Schematic of the DNN structure

x1
x2 y1
x3 y2
…
…

xn ym

Input layer Hidden layer Output layer

Figure 2:

Title: Wave velocity vs.SPof sandstone and mudstone in the target area

Legend: Figure 2 show Wave velocity vs.SPof sandstone and mudstone in the target area

14
 Sandstone
5000
Mudstone

4500

Wave velocity(m/s) 4000

3500

3000
(m/s )

2500

2000
-10 0 10 20 30 40 50 60 70

Spontaneous-potential(mv)

Figure 3:

Title: SP curves of some logsinvolved in multi-attribute inversion (after normalization)

Legend: Figure 3 show SP curves of some logsinvolved in multi-attribute inversion (after

normalization)

1400 1600 1600 2000

1600 2200
1800
1800
1800 2400
2000
2000 2600
2000
2200
Depth(m)

2200 2800
2200
2400
(m)

2400 3000
2600 2400
2600 3200

2800
2800 2600 3400

3000 3600
3000
2800
3200 3200 3800

3400 3400 3000 4000

0 0.5 1 0 0.5 1 0 0.5 1 0 0.5 1
SP SP SP SP
(a)Niu103 (b) Niu26 (c) Wang580 (d) Wang78

15
Figure 4:

Title: Quantity of seismic attributesvs. error

Legend: Figure 4 show Quantity of seismic attributesvs. error

Training error
0.24
Testing error

0.22
Error(mv)

0.20

0.18
1 2 3 4 5 6 7 8
Quantity of seismic attributes

Figure 5:

Title: Well section of the intermediate section in Member 3 of the Shahejie Formation in the

Dongying Delta as obtained by multi-attribute SP inversion

Legend: Figure 5 show Well section of the intermediate section in Member 3 of the Shahejie
Formation in the Dongying Delta as obtained by multi-attribute SP inversion

16
 CDP
3100 3500 3900 4280
1200

Niu 26

Niu 80
Wang78

Delta plain
Times(ms)

2000

Delta front

Front delta

2800

Figure 6:

Title: Horizon slice of the intermediate section in Member 3 of the Shahejie Formation in the
Dongying Delta as obtained by multi-attribute SP inversion

Legend: Figure 6 show Horizon slice of the intermediate section in Member 3 of the Shahejie

Formation in the Dongying Delta as obtained by multi-attribute SP inversion

17
 Horizontal line

0.9
The
z7

0.7

Spontaneous-potential
Foreset
Vertical
line

z5

0.5

Sand

z3

0.3

Body

z1

18