0 Voti positivi0 Voti negativi

4 visualizzazioni5 pagine4D wavelet estimation

May 23, 2016

SEG-2014-0667

© © All Rights Reserved

PDF, TXT o leggi online da Scribd

4D wavelet estimation

© All Rights Reserved

4 visualizzazioni

SEG-2014-0667

4D wavelet estimation

© All Rights Reserved

- Co 33548550
- steidl-sinum04.pdf
- WAVELET1
- IEEE SP Sukittanon03
- Data Processing of Resistivity Survey
- Using Priority Curve Algorithm in Sports Video Summarization
- 1-s2.0-S1385894703002444-main.pdf
- DECS Finalised
- Enhancement of Underwater Images by Image Fusion Using Wavelet Decomposition
- Audio Watermarking in Image by Using Radon – Wavelet Transforms
- Seismic Data Acquisition
- Resonance Paper1 Mar2004
- AC motors
- West et al 2002 TLE
- Wavelet
- Automatic Fault Detection for 3D Seismic Data
- IJETTCS-2017-03-01-2
- 2006 - A Quantitative Identification Approach
- CV (Mahboob)1 PDF
- 00881961

Sei sulla pagina 1di 5

Summary

In this paper we propose a new approach to estimate

wavelets associated with 4D data. Our approach does not

use well logs and therefore can estimate wavelets at various

space positions and so enables the estimation of laterally

varying wavelets. We assume that 4D signal solely occurs

at a limited number of layers. Using this simple observation

it is possible to invert for reflectivity series and wavelets at

the same time. We present results obtained on real data and

compare them to traditional estimated wavelets using welllogs.

Introduction

Wavelet estimation is still a key question in geophysics.

Technologies used for wavelets estimation relate on the

convolution equation (Walden and White 1984):

() = () () + () (1).

Where S(t) represent the seismic signal, r(t) the reflectivity

series of the earth, w(t) the wavelet and n(t) the noise.

Usually the following assumptions are made: noise is

uncorrelated and has a gaussian distribution (always), and

reflectivity series is white (often). Based on these

assumptions common approaches for wavelet estimation

include:

statistics of traces around the well, find the best

trace position and finally estimate the phase

which provides the best match between seismic

data and synthetics (White and Simm 1983). This

strategy can also be applied to an ensemble of

wells as in Malkin and Canning( 2008).

spectrum from statistics of traces, then estimate

the phase using high order statistics (Edgar and

Van der Baan 2011). Another approach is to

assume that wavelet has a minimum phase which

therefore could be solved analytically (Brown et

al 1988).

not exact. Bayesian approaches have been

developed (Buland and Omre 2003 and Gunning

and Glimsky2006) that introduce uncertainties

such those related to inaccuracy in time to depth

conversions,.

reflectivity series and using linear programming

2014 SEG

SEG Denver 2014 Annual Meeting

altogether (Yao and Galbraith 2012).

Each of these approaches has advantages and

disadvantages:

The use of wells is advantageous in the sense that it

enables: 1 - to estimate the true amplitude of the wavelet,

2 - to correct the spectrum of the wavelet (reflectivity series

are not white), and 3 - to estimate the phase without

additional assumptions. This method also has clear

shortcomings in the sense that it is only feasible at well

position. This limitation has several implications: Wells are

sparse and hence assumptions must then be made about the

stationarity of the wavelet; very few wells drilled are

vertical and tying to the well in this situation is

problematic.

Bayesian approaches tend to handle the inaccuracy of the

reflectivity series but results are strongly dependant on

assumptions made about those uncertainties. Besides, most

seismic processes and inversions accept only a single

realization of the wavelet even though it is possible the

main source of uncertainty in an inversion (Gunning and

Glimsky 2006, Thore et al 2013).

Use of seismic data only is obviously a big advantage in the

sense that it gets rid of the well, associated uncertainties

and spatial limitations. On the other hand high order

statistics are very geologically driven (Xu et al 2012) and

phase estimation strongly unreliable. Sparse deconvolution

approach is extremely sensitive to noise and strongly

dependant on the relative weights between fitness of the

data versus sparseness of reflectivity series.

The techniques that we mentioned have been developed for

3D data and are applied straightforwardly to 4D data: 4D

processing aims at providing equivalent datasets for various

vintages and is generally well achieved by matching areas

outside where significant 4D signal exists. The spectrum

and phase of the various vintages are matched and so the

wavelet calculated with the base dataset could be

adequately used for the other vintages. Unfortunately this

assumption is not always true and due to many factors

(mainly acquisition) it is not always possible to match

everywhere the different vintages. In addition to this

problem it is often necessary to invert two monitors for

which no sonic and density logs have been acquired and so

the wavelet estimation techniques using well are

inoperable. Of course processor wizards will always found

a solution and provide wavelets that will make the job. In

this paper we propose a new way to estimate wavelet

dedicated to 4D data which does not make use of well data.

DOI http://dx.doi.org/10.1190/segam2014-0667.1

Page 4878

4D Wavelet Estimation

Method

Thore and Hubans (1992) have proposed a method for

estimating 4D signal at a well. Their algorithm is based on

the assumption that flows during production occur in a

layered manner and that inversion could be done on a

blocky model (figure 1). We use the same assumption for

wavelet estimation: we assume that dynamic flows are

confined into a limited number of layers. Therefore we are

in the situation that is presented on the figure 1.

() () = ( ) ()

(2)

In equation (2) () is warped Monitor trace, () the

flow simulation time of top reservoir time of

bottom reservoir, () the wavelet, () represents the

Dirac function and the convolution.

In equation (2) the left hand side is known fairly accurately

(provide that the Monitor trace has been correctly aligned

to the Base) and the unknowns are: , the position of

the of the top and the base and the wavelet (). Positions

of the top and base of the reservoir can be estimated

approximately and is just a scaling factor.

Equation (2) can easily be extended to a set of reservoir

layers:

() () =

() (3)

conduct to the final function to minimize:

= ( () ())

,()2 (4)

acquired before production. Track 2: 4D perturbation

due to production confined in one single layer. Track 3:

Base and Monitor traces superimposed. The Monitor

trace differs from the base by to aspects: first time shift

below reservoir and amplitude at top and bottom of the

reservoir. Track 4: Base trace and Monitor trace

corrected of time shift effect.

For the sake of simplification we only consider the case

where all the flow occurs in a single layer. The Monitor

after time alignment to the Base trace differs from the Base

by an amplitude change at the top and at the base of the

reservoir layer. Those amplitude changes are opposite sign

and with a good approximation can be considered to be of

same magnitude. Whatever the impedance of the overburden and of the under-burden the variation of reflection

coefficient can be expressed as:

1

1

2

(1)

of the true reflection coefficient, R.

Therefore the difference between Base trace and warped

Monitor trace it can be expressed as:

2014 SEG

SEG Denver 2014 Annual Meeting

As mentioned earlier all the unknowns are not totally

independent: the positions of the top and base and the

reflection coefficient of each layer are correlated from one

trace to the next (as are the traces) and the positions of the

top and base of the layers are known approximately.

Using equation (5) it is impossible to solve for the absolute

value of the wavelet since there is clear tradeoff between

wavelet amplitude and reflection coefficient magnitude.

The simplest way to remove this ambiguity is to fix the

maximum amplitude of the wavelet to a reasonable value.

There is also a tradeoff between a constant time lag of the

wavelet and the reflector position. We have chosen to

constrain the wavelet to have zero time lag.

The total number of unknowns is:

= (1 + 2) + (5)

Where is the number of layers, the number of traces

and the number of samples of the wavelet.

It is clear that this inverse problem is solvable only if is

small compare to the total number of samples, i.e. if is

relatively small.

The wavelet could have been estimated in the spectrumphase domain: getting the spectrum from inter-correlation

of seismic data and assuming some type of phase behavior

such as constant phase; It is exactly what we want to avoid:

DOI http://dx.doi.org/10.1190/segam2014-0667.1

Page 4879

4D Wavelet Estimation

convolution of the wavelet by a few well individualized

reflection coefficients and so its spectrum will be affected

by a set of notches..

We have decided to use a global optimisation algorithm to

allow for the largest variations of the wavelet. Our choice is

CMAES (Covariance Matrix Adaptative Evolution Strategy

(Auger and Hansen 2011)) which has been shown to be

extremely efficient to resolve geophysical problem (Thore

2012) . The technique uses an ensemble of models and at

each iteration it computes a gaussian approximation of the

data space which is then used to generate a new ensemble

of models (see Figure 2).

Although the starting point is theoritycally not an issue

with this algorithm, starting with a reasonnable set of

models is always benficial. As we have mentionned above

we can provide a rough appromixation of the position of

the reflectors with a picking of the 4D anomalies. For the

reflection coefficients we just use a series of [1,-1]. For the

wavelet we use a sinc function based on the maximal

frequency measured on the seismic data.

The number of traces that we use depends on the noise

level and the type of structure but an average of 100 traces

is a good guess.

The convergence is quite fast: around a thousand of

generations are required with a population size of 50

individuals. Altogether the estimation is performed in a few

seconds once the layer definition is done.

used for the wavelet estimation.

used for the seismic data and the picking of the top

reservoir.

Figure 4 shows the synthetic seismic the residuals and the

reflectivity estimated during the inversion while Figure 5

shows the comparison of the estimated wavelet using the

4D data and the one estimated using the nearby well.

ensemble of models at various generations.

The case presented here is based on an application in a field

of the Gulf of Guinea where signal to noise ratio is quite

high and where the seismic to well tying is good (therefore

we will have a good reference wavelet).

2014 SEG

SEG Denver 2014 Annual Meeting

procedure. Top: comparison of the real and synthetic

seismic for one of model provided by the inversion

algorithm. Bottom: residuals (real data- synthetics)

and reflectivity coefficients estimated as a side

product of the inversion.

DOI http://dx.doi.org/10.1190/segam2014-0667.1

Page 4880

4D Wavelet Estimation

Conclusions:

We propose a new technology for estimating wavelets

using 4D data but without using wells. This methodology is

simple, efficient and requires little human interaction. Our

approach uses a global search algorithm and therefore

requires very little prior information. This methodology can

be applied at various places to check the stability of the

wavelet. It provides optimal wavelets to be used in 4D

inversion.

only 4D data (green) and using Base seismic and a

nearby well (red) using the 3D data.

The results are excellent and the wavelet estimated with

solely the 4D data is quite close to the one estimated using

the Base seismic and the nearby well. The 4D-wavelet

shows no oscillation although the initial solution was a sinc

function and no constraint was introduced between the

samples during the inversion.

Spatially varying wavelet:

In this section we test the stability of the wavelet by

running the inversion at different location using different

4D anomalies. Figure 6 sum up the results that we have

obtained. The wavelet has been estimated at 5 positions

labeled A, to E. Positions A to C (Figure 6) are all within

homogeneous water pools and the corresponding wavelets

are quite similar. Wavelet D was estimated in the same

pool area as A and B but several traces used in the

inversion contained stacked 4D anomalies (2 layers),

nevertheless the estimated wavelet is still clean and very

much alike wavelets A, B and C. Finally wavelet E was

estimated in an area where the 4D signal is comprised of

multiple layers (>3) and complex production mechanisms

exist in the same location. As a consequence, the estimated

wavelet shows large oscillations. If one considers only

wavelets A to D the results of the 4D wavelet inversion are

very consistent and the variability observed is not larger

than the one that is observed when estimating wavelets

using different wells. Is this variability real or is it inherent

to the inversion process we cannot currently answer this

question.

2014 SEG

SEG Denver 2014 Annual Meeting

location. Wavelets A, B, and C have been extracted in

water injected pool, wavelet D was obtained in the

water pool but slightly overlapping on a gas pool and

wavelet E was obtained in a mixed area with very

poor 4D signal.

Acknowledgements:

The authors would like to thank Erwan Letard who has

worked on this project during an internship at the GRC.

DOI http://dx.doi.org/10.1190/segam2014-0667.1

Page 4881

http://dx.doi.org/10.1190/segam2014-0667.1

EDITED REFERENCES

Note: This reference list is a copy-edited version of the reference list submitted by the author. Reference lists for the 2014

SEG Technical Program Expanded Abstracts have been copy edited so that references provided with the online metadata for

each paper will achieve a high degree of linking to cited sources that appear on the Web.

REFERENCES

Auger, A., and N. Hansen, 2011, Tutorial: CMA-ES Evolution strategies and covariance matrix

adaptation: www.lri.fr/~hansen/gecco2011-CMA-ES-tutorial.pdf.

Brown, R. L., W. McElhattan, and D. J. Santiago, 1988, Wavelet estimation: An interpretive approach:

The Leading Edge, 7, 1619, http://dx.doi.org/10.1190/1.1439470.

Buland, A., and H. Omre, 2003, Bayesian wavelet estimation from seismic and well data: Geophysics, 68,

20002009, http://dx.doi.org/10.1190/1.1635053.

Edgar, J., and M. van der Baan, 2011, How reliable is statistical wavelet estimation?: Geophysics, 76, no.

4, V59V68, http://dx.doi.org/10.1190/1.3587220.

Gunning, J., and M. Glinsky, 2006, WaveletExtractor: A Bayesian well-tie and wavelet extraction

program: Computers & Geosciences, 32, no. 5, 681695,

http://dx.doi.org/10.1016/j.cageo.2005.10.001.

Malkin, A., and A. Canning, 2008, Global multiwell wavelet estimation: 78th Annual International

Meeting, SEG, Expanded Abstracts, 20262030, http://dx.doi.org/10.1190/1.3059290.

Thore, P., 2012, Solving the multimodal problem of 4D seismic to well tying, a comparison of three

different stochastic algorithms: 82nd Annual international Meeting, SEG, Expanded Abstracts,

doi:10.1190/segam2012-0122.1.

Thore, P., and C. Hubans, 2012, 4D seismic -to-well tying, a key step towards 4D inversion: Geophysics,

77, no. 6, R227R238, http://dx.doi.org/10.1190/geo2011-0267.1.

Thore, P., Y. Xu, and O. Duplantier, 2013, Seismic inversion uncertainty: What does really matter?: 83rd

Annual international Meeting, SEG, Expanded Abstracts, 30883092,

http://dx.doi.org/10.1190/segam2013-0174.1.

Walden, A. T., and R. E. White, 1984, On errors of fit and accuracy in matching synthetic seismograms

and seismic traces: Geophysical Prospecting, 32, no. 5, 871891, http://dx.doi.org/10.1111/j.13652478.1984.tb00744.x.

White, R. E., and R. Simm, 2003, Tutorial: Good practice in well ties: First Break, 21, 7583.

Xu, Y., P. Thore, and S. Duchenne, 2012, The reliability of the kurtosis-based wavelet estimation: 82nd

Annual international Meeting, SEG, Expanded Abstracts, http://dx.doi.org/10.1190/segam20121221.1.

Yao, Z., and M. Galbraith, 2012, Seismic source-wavelet estimation and sparse spike deconvolution: 82nd

Annual international Meeting, SEG, Expanded Abstracts, http://dx.doi.org/10.1190/segam20121409.1.

2014 SEG

SEG Denver 2014 Annual Meeting

DOI http://dx.doi.org/10.1190/segam2014-0667.1

Page 4882

- Co 33548550Caricato daShakeel Rana
- steidl-sinum04.pdfCaricato daehsann
- WAVELET1Caricato day2012asan
- IEEE SP Sukittanon03Caricato daakoca23
- Data Processing of Resistivity SurveyCaricato damkdevarajan
- Using Priority Curve Algorithm in Sports Video SummarizationCaricato daIJSRP ORG
- 1-s2.0-S1385894703002444-main.pdfCaricato daMuhammad Adnan Laghari
- DECS FinalisedCaricato damurthykalki
- Enhancement of Underwater Images by Image Fusion Using Wavelet DecompositionCaricato daijsret
- Audio Watermarking in Image by Using Radon – Wavelet TransformsCaricato daJournal of Telecommunications
- Seismic Data AcquisitionCaricato daRofeeha Atinuke Olagunju
- Resonance Paper1 Mar2004Caricato daSneha Nanavati
- AC motorsCaricato daSyedshabab Haider
- West et al 2002 TLECaricato daPatrick Ellis
- WaveletCaricato dastimoceiver
- Automatic Fault Detection for 3D Seismic DataCaricato daPedro Mureb
- IJETTCS-2017-03-01-2Caricato daAnonymous vQrJlEN
- 2006 - A Quantitative Identification ApproachCaricato dapaulkohan
- CV (Mahboob)1 PDFCaricato daShahroz Noumani
- 00881961Caricato daapi-3706534
- Satellite Image Classification Using Wavelet TransformCaricato daIAEME Publication
- Petroleum 00 BishCaricato daManthneha
- Lifting based 3D Discrete Wavelet Transform for Image CompressionCaricato daIJRASETPublications
- 9781447149408-c2Caricato daFabiola Chacha
- Image Denoising of various images Using Wavelet Transform and Thresholding TechniquesCaricato daAnonymous CUPykm6DZ
- najafabadian2018Caricato daSaid Djaballah
- Image CompressionCaricato daMeenusruthi
- Research InitiativeCaricato darupesh
- laplace derivatifCaricato daMuhammad Iqbal
- Digital Image ReconstructionCaricato daulfaulfa

- Seismic MigrationCaricato daniksr91
- 4d Time Lapse ImagingCaricato daniksr91
- Spectral DecompositionCaricato daniksr91
- Carbonate Reservoir Evaluation Using Rock PhysicsCaricato daniksr91
- image processingCaricato daniksr91
- Attribute AnalysisCaricato daniksr91
- enhanced delineation of reservoir compartmentalization from advanced pre and post stack seismic attribute analysisCaricato daniksr91
- Dissertation Report- Pranshu RatreCaricato daniksr91

- PDPIPIDCompCaricato daDin_rx
- Lect Notes 8Caricato daSafis Hajjouz
- GLCMCaricato daprotogizi
- LZMWCaricato daNinggar Anggarwati
- Chaos-Classical and quantumCaricato dabharaths100
- An Improved Genetic Algorithm for the Vehicle Routing Problem WithCaricato dareubendemetrius
- Speech processing research paper 1Caricato daimparivesh
- Ensembling test.docxCaricato dajecko augustine
- A Survey on Clustering Techniques in Medical DiagnosisCaricato daEighthSenseGroup
- Introduction to dependent type theory by Yves LafontCaricato daTor Økland Barstad
- imageCaricato davishwanath
- 2012-13Caricato daabhishek
- Approximation of Pi Using the Monte Carlo MethodCaricato daapi-3798769
- What is the Easiest Definition of 'Entropy' - QuoraCaricato daMark J. Burton II
- 10 Mathematics Ncert Ch01 Real Numbers Ex 1.1Caricato daakkmt111
- Chapter 1 ErrorsCaricato dataqiyuddinm
- Implementing SSLCaricato daSavas Kaplan
- Cv Ahmad El-AjouCaricato daDr-Ahmad El Ajou
- Bouvet.pdfCaricato daAhmed58seribegawan
- Chapter 10Caricato daJem Mallari
- Bayesian Approach for Animal Breeding Data AnalysisCaricato daGopal Gowane
- Simple FftCaricato daSudeshna Kundu
- Image and Video Super-resolutionCaricato dajebemnato
- BITS F111 Thermodynamics Handout 2014-15Caricato dashivaraj1996
- An Introduction to Numerical Integration or QuadratureCaricato darodwellhead
- Simulation in terminated systemsCaricato daSaleem Almaqashi
- Lecture 1.pptxCaricato daAbay Yeshaw
- MS(Transp.transship,Assign) 1Caricato daAbhishek Sharma
- AnswersCaricato daOc Loi
- SolutionAzizHW2Caricato daSifun Padhi

## Molto più che documenti.

Scopri tutto ciò che Scribd ha da offrire, inclusi libri e audiolibri dei maggiori editori.

Annulla in qualsiasi momento.