Download as pdf or txt
Download as pdf or txt
You are on page 1of 35

FINAL YEAR PROJECT II

Sand Volume Prediction Based


on Elastic Properties using
Neural Network
STUDENT : NIK MUHAMMAD IZZAT BIN MOHD ZAINI
MATRIC ID : 20871
SUPERVISOR : DR MAMAN HERMANA
INTRODUCTION
PROBLEM STATEMENT
OBJECTIVES
SCOPE OF STUDY

OUTLINE LITERATURE REVIEW


METHODOLOGY
PROJECT MILESTONE AND TIMELINES
RESULT AND DISCUSSION
CONCLUSIONS AND RECOMMENDATIONS
INTRODUCTION
• Accurate prediction of lithology and pore fluid is very essential and acts as a key
challenge for prospective hydrocarbon exploration and development.

• Sand distribution identification is important to identify the potential reservoir area


that hydrocarbon can be accumulated.

• Artificial neural network (ANN) has been used to model single or multiple target
properties from predictor variables in different research problems because of its
accurate prediction and generalization capability.
PROBLEM STATEMENT
• In common method, sand lithology is predicted from GR response and it is cover
only at well location.
• Seismic data cover the large area but only provides the interface properties not the
layer properties.
• Lithology prediction especially sand volume distribution from elastic properties is
an important task need to be done. Limitation on the estimation result on current
technology will try to be solved by including neural network and new attributes
developed in UTP.
OBJECTIVES

1. To classify lithology using elastic properties on well log.

2. To investigate the sensitivity of elastic properties due to sand


volume changes.

3. To implement Artificial Neural Network (ANN) for 3D sand


distribution.
SCOPE OF STUDY
• Well-log study for lithology analysis

• Rock Physics study for elastic behaviour

• Application of Artificial Neural Network for sand volume


prediction

• 3-D SEG-Y sand volume prediction across the reservoir in


hydrocarbon prediction.
LITERATURE REVIEW
Artificial Neural Network (ANN)
• Russel (2013) clarified few stages in sample-to-sample prediction.
• Additionally, it is likewise restricted to stochastic strategies, which convey different
potential models that are also plausible.
• The primary stage in sample-to-sample prediction is training the neural network or
multilinear regression at the wells.
• Next, to build up a parameter of interest, a series of attributes are selected.
• Lastly, the output is contrasted with known wells to find errors, which are then fed
back through the system to decide a superior weight for each property.
Artificial Neural Network (ANN)

Figure 1: Artificial Neural Network Model


retrieved (Russel, 2013)

Figure 2: Example of good predicted porosity with


actual porosity (Verma, 2012)
Simultaneous Seismic Inversion
• The conversion of seismic data to elastic
properties through inversion of multiple
partial angle stack
• More than one wavelet where they are
extracted from each point stack
• Seismic data and well data will both be
used to generate the inversion model.
• This would be a significant result which
shows logs and reflection seismic data can Figure 3: Simultaneous inversion
(wiki.seg.org/wiki/Seismic_inversion)
be used as an additional formation
evaluation tool to quantify the distribution
of sand in a reservoir.
SQp and SQs for lithology and fluid indicator
• New lithology and fluid indicator called SQp and SQs were developed from seismic attenuation
rock physics approximation. The attenuation and phase velocity of plane wave propagation in the
viscoelastic medium is governed by Kramers-Kronig relation and maximum of Quality factor of P
wave (Qp) and S wave (Qs) can be estimated using the following equations:
𝑀 −𝑀
• 2𝑄𝑝−1 = ∞ 0
𝑀𝑜 𝑀∞
𝐺∞ − 𝐺0
• 2𝑄𝑠 −1 =
𝐺𝑜 𝐺∞
• This equation shows the inverse of the quality factor is estimated from high frequency and low
frequency situations. However, direct prediction from seismic data is complicated. The high and
low frequency condition in rock physics approximation can be assumed as an effect of crack based
on Hudson Crack Theory (Hermana et al.2017).
2
𝑉𝑝 2
5 1 ( 𝑉𝑠 ) −2
• 𝑆𝑄𝑝−1 =
6  𝑉𝑝
(( 𝑉𝑠 )2 −1)
𝑉𝑝
10 1 ( )2
• 𝑆𝑄𝑠 −1 = 𝑉𝑠
3  𝑉𝑝 2
(3( 𝑉𝑠 ) −2)
SQp and SQs response on well log

Figure 4: SQp and SQs logs compared with NPHI, Vclay, and Resistivity
Logs (Hermana et al. ,2017).
PROJECT WORKFLOW
Seismic Stacking
Well-log data
(Near, Mid, Far) Elastic Properties attributes formula
• P-Impedance : AI = Vp x ρ , where Vp: P-wave,
Well-log analysis Simultaneous Inversion ρ: density
• S-Impedance : SI = Vs x ρ , where Vs: S-wave, ρ:
Elastic properties attributes
density
calculation on well log Output: ρ,Vp,Vs
interpretation • Lambda-Rho,  = 𝐴𝐼 2 − 2𝑆𝐼 2 where AI: P-
Impedance, SI: S-Impedance
Attributes Application (SQp, SQs, • Mu-Rho, μρ = SI² where SI: S-Impedance
Cross-plot
SQp/SQs) • SQp = (5/6) *(1/Rho) *((a-2)^2)/(a-1)) where a:
(Vp/Vs)^2
• SQs = (10/3) *(1/Rho) *((a)/((3*a)-2)) where a:
Artifical Neural Network (Vp/Vs)^2

3D Sand Distribution Modelling


MILESTONES

FYP 1 & 2 GANTT CHART & MILESTONES Project planning and data
gathering (15 June 2018)

Well Log analysis (13 July


2018)

Lithology classification (20


July 2018)

Elastic Properties Cross


Plotting (10 Aug 2018)

Simultaneous seismic
inversion (14 Sept 2018)

Attribute analysis (12 Oct


2018)

FYP 1 & 2 GANTT CHART Applying Artificial Network


modelling (29 Oct 2018)

3D Sand volume distribution


modelling (9 Nov 2018)
RESULT AND
DISCUSSION
Location Map

Figure 6: Location map of X field


Topography Map

SS1

A
Well Log interpretation
Elastic Properties
Cross-plot

Figure 11: Vsand vs P-Impedance cross-plot Figure 13: Vsand vs Lamda-Rho cross-plot

Figure 12: Vsand vs S-Impedance cross-plot Figure 14: Vsand vs Mu-Rho cross-plot
Cross-plot
Figure 6: Summary of cross correlation of elastic
properties with sand volume

Input Equation Correlation


Coefficient
P-Impedance Vsand =-0.000553 (AI) +100.29 0.2324

S-Impedance Vsand = -0.017 (SI) + 106.265 0.3730


Figure 15: Vsand vs SQp cross-plot
Lambda-Rho Vsand = -1.27 () + 81.65 0.5436

Mu-Rho Vsand = -1.89368 (µ) + 89.27 0.3182


SQp Vsand = -57.0319 (SQp) +77.55 0.6981
SQs Vsand=119.926(SQs) –13.3153 0.8661

Figure 16: Vsand vs SQs cross-plot


Simultaneous Seismic Inversion

Figure 17: Inversion result for Density, P-Impedance, Vp/Vs


Simultaneous Seismic Inversion
SQp
(unitless)

Figure 18: Inversion result SQp


Simultaneous Seismic Inversion
SQs
(unitless)

Figure 19: Inversion result SQs


Direct prediction using cross-plot equation
Volumetric Volumetric
(%) (%)

Figure 20: Sand Volume prediction using direct cross-plot equation


Artificial Neural Network

Figure 22: Predicted Figure 23: Predicted


Figure 21: Multiple attribute data input Vsand using 4 attributes Vsand using 6 attributes
Artificial Neural Network

Figure 24: Vsand actual vs Vsand predicted


Comparison Neural network and direct model
Color Data: Vsand_from_ANN
Inserted Data: Volumetric

Figure 26: Sand volume prediction ANN Figure 27: Sand volume prediction using direct cross-plot
equation
Sand volume prediction in 3-D

Figure 28: Sand volume prediction in 3-D


Sand volume prediction distribution using
Neural Network

Figure 28: Sand volume prediction from top view


CONCLUSIONS AND
RECOMMENDATIONS
Conclusions
1. Integrating and calibrating the data from well-log and seismic would
ease the prediction the lithology within a formation and making the
results more reliable.
2. The new attribute namely SQp and SQs which are derived from
attenuation by using rock physics approximation give some significant
improvement not only in determining the lithology but also in
predicting sand volume.
3. Direct sand volume prediction using SQs attribute also give significant
result in predicting sand volume.
4. Neural network provide clear image to demonstrate the sand volume
distribution as compared to direct estimation.
Recommendations
• Recommended to have core samples and data from various fields
with various depositional environment background
• Different Neural Network structure can be utilized alternatively in any
studies.
• It is recommended to use the developed Artificial Neural Network
model in integrating the results of lithology and fluid content
prediction.
References
Bagheripour, P. (2014). Committee neural network model for rock permeability prediction. Journal of Applied Geophysics, 104,
142-148. doi:10.1016/j.jappgeo.2014.03.001

Berge, T. B., Aminzadeh, F., de Groot, P., & Oldenziel, T. (2002). Seismic inversion successfully predicts reservoir, porosity, and
gas content in Ibhubesi Field, Orange Basin, South Africa. The Leading Edge, 21(4), 338-348.

Hermana, M., Ghosh, D. P., Sum, C. W., & Salim, A. M. A. (2016, November). Implementation of New Seismic Attributes to
Improve Reservoir Properties Prediction Using Probability Neural Network. In International Petroleum Technology Conference.
International Petroleum Technology Conference.

Hermana, M., Ngui, J. Q., Weng Sum, C., & Prasad Ghosh, D. (2018). Feasibility Study of SQp and SQs Attributes Application
for Facies Classification. Geosciences, 8(1), 10.

Russell, B. (2013, Janurary). Neural Networks find meaning in data . Retrieved from from
https://www.cgg.com/technicalDocuments/cggv_0000016502.pdf

Verma, A. C. (2014). Porosity and permeability estimation using neural network approach from well log data . American
Association of Petroleum Geologists.
Thank You

You might also like