Integrated Reservoir Characterization for Unconventional ...€¦ · evaluation and manual training...

Integrated Reservoir Characterization for

Unconventional Reservoirs using Seismic,

Microseismic and Well Log Data

Debotyam Maity

PhD Defense

Advisor: Fred Aminzadeh

June 13, 2013

06/13/2013 PhD Defense 2

Outline

Introduction

Advanced multiphase autopicker design and implementation workflow

Attribute studies based on conventional seismic data

Passive seismic based geomechanical property estimation

Hybrid FZI* attributes

Integrated analysis

Conclusions

* Fracture Zone Identifier

Part 1

Introduction

Problem definition

Background

Workflow


Background

Unconventional Resources?

Shale Geothermal

CO2

injectionSteam

flooding

Water flooding

Gas flooding

Increasing stress on conventional oil resources has compelled us to look intounconventional oil & gas and other subsurface energy resources.

Introduction AutopickingGeomechanical

property estimation

Seismic attribute studies

FZI attributes Integrated analysis Conclusions

Background

Geophysical tools employed…

Seismic and microseismic surveys have become increasingly common. Loweringacquisition costs, improvements in capabilities and new techniques for improvedreservoir characterization have all led to the abundance of such geophysical studies.

Reflection seismology

(Nissen, 2007)

Passive seismology

(Calvez, 2005; ESG Solutions)


property estimation



Background

Anjaneyulu et al., 2011

1. Seismic attributes used to characterize subsurface channels2. Seismic attributes used to characterize fractures & other discontinuities3. Results of “ant-tracking” algorithm with traditional attributes

Li, 2007

4. AVAZ analysis on pre-stack data for fracture characterization

Basir et al., 2013

Background

1. MEQ monitoring in Barnett shale – hydraulic fracture characterization

Urbancic et al., 2002

2. Edge detection to detect potential fault reactivation or fault barrier

Maxwell et al., 2011

Cipolla et al., 2012

3. Integrated evaluation using stress anisotropy, seismic curvature & seismic fabric direction used to determine fracture zone complexity

4. Identifying fault activation in Barnett using spatial b-value mapping (and energy release)

Cipolla et al., 2012

5. Complex hydraulic fracture geometry compared with MEQ cloud and modeled 20 year pressure depletion for un-propped zones

Wessels et al., 2011

Study area


property estimation



Geothermal Sedimentary Fluvial deltaic Faulted tectonic Compartmentalized

Problem definition

Motivation?

Enhanced cross-disciplinary technology applications.

How to work in highly data constrained & geologically challengingenvironments?

Novel workflows to tackle said challenges.

Maximize/ optimize use of available data.

Improved algorithms to support analysis.

Improved velocity models

Fracture zone characterization

Discontinuity mapping

4D characterization framework

MEQ – Seismic joint interpretation


property estimation



Integrated characterization workflow

Passive seismic

3D seismic

Well Logs

Data conditioning

Dip steered filtering

Well to seismic ties

Seismic attribute analysis

Multi-attribute/

ANN

Reservoir property estimates

Rock properties

A-priori information on fracture zones

Data formatting

Auto-picking

Phase detection

Event locations

Tomographic inversion

COSGSIM

Vp & Vs (high resolution)

ANN classification

algorithm

FZI maps

Image logs/ production data,

etc.

Estimation uncertainty

Inversion uncertainty

Fracture zone identification

framework

Part 2

Advanced multiphase autopicker

Background

Motivation

Workflow

Implementation

Results


Background

To detect onset of phase energy arrivals

P phase

S1 phase

S2 phase

Available methods based on previous work

STA/LTAAllen, 1978

Energy changeCoppens, 1985

Wavelet transformationKanwaldip et al., 1978

AICMaeda et al., 1985

Sleeman et al., 1999

Leonard et al., 1999

Skewness/ KurtosisSaragiotis et al., 2002

ANNZhao et al., 1999

FractalBoschetti et al., 1996

Chang et al., 2002

Cross-correlationRowe et al., 2002

Begnaud et al., 2007

StatisticsWagner et al., 1996

06/13/2013 PhD Defense 12Introduction AutopickingGeomechanical

property estimation




Initial observations

Random noise

Dead/ near dead traces

High frequency harmonics

Low frequency artifacts

Scattered energy

Which pick to choose? What criteria serves as the best?


property estimation



AIC & STA/LTA methods

Motivation

5 recording stations

Improved pick accuracy

Ability to handle untested data

Quick phase data generation for near

real time applications

Hydrofrac/ Geothermal, etc

ANN autopicker workflow

DMX files Stacked dataset

Rotate data to have maximized p phase energy

projection on vertical component

Apply bandpass filter based on frequency

spectrum of individual traces

Attribute selection based on literature/ cross-plotting & window

size analysis (sensitivity studies).

Manual picks for ANN training and validation

ANN training using selected attribute maps

Iterate to improve attribute selection and reduce total no. of attributes

Extract ANN picks from the obtained ANN probability map

Preliminary comparisons with contemporary autopickers and

expert validation

Extract seismic trace values starting from p phase arrivals (as obtained from first run of

the autopicker) : 2 s time window

horizontals vertical1

1

Apply high pass filter to attenuate low values close to p phase pick

(obtained automatically from frequency spectrum)

ANN training using previously obtained attribute maps (p

phase trained ANN)

New attributesiterate

Extract ANN picks from the obtained ANN probability

map – s phase arrival offsets

Obtain final pick locations (p and s phase). Compare results and

calibrate workflow as required

Theory - ANN

𝐀𝟏𝐢 =

𝐢−𝐍

𝐢+𝐍

𝐱𝐢

AbsSum

𝐀𝟐𝐢 =𝟏

𝐍

𝐢−𝐍𝟐

𝐢+𝐍𝟐

𝐀𝟏𝐢

AbsSumMean

𝐀𝟑𝐢 = 𝐦𝐚𝐱(

𝐢−𝐍𝟐

𝐢+𝐍𝟐

𝐀𝟏𝐢)

AbsSumMax

𝐀𝟒𝐢 = 𝐯𝐚𝐫(

𝐢−𝐍/𝟐

𝐢+𝐍/𝟐

𝐀𝟏𝐢)

AbsSumVar

𝐀𝟓𝐢 =

𝟏𝐍×

𝐢−𝐍𝟐

𝐢+𝐍𝟐 𝐱𝐢 − 𝐱 𝟑

(𝟏𝐍×

𝐢−𝐍𝟐

𝐢+𝐍𝟐 𝐱𝐢 − 𝐱 𝟐)

𝟑𝟐

Skewness

𝐀𝟔𝐢 = −𝟑 +

𝟏𝐍×

𝐢−𝐍𝟐

𝐢+𝐍𝟐 𝐱𝐢 − 𝐱 𝟒

(𝟏𝐍×

𝐢−𝐍𝟐

𝐢+𝐍𝟐 𝐱𝐢 − 𝐱 𝟐)

𝟐

Kurtosis

𝐀𝟕𝐢 = 𝐍 × 𝐥𝐧(𝟏

𝐍×

𝐢−𝐍𝟐

𝐢+𝐍𝟐

𝐱𝐢 − 𝐱 𝟐) + 𝐤 × 𝐥𝐧(𝐍)

BIC

𝐀𝟖𝐢 = 𝐭𝐚𝐧−𝟏𝐇 𝐭

𝐱 𝐭

InstPha

𝐀𝟗𝐢 =𝐝

𝐝𝐭𝐀𝟖𝐢

InstFreq

𝐀𝟏𝟎𝐢 =𝟏

𝐥𝐨𝐠𝐅× 𝐥𝐨𝐠

𝐦𝐚𝐱 ∆𝐱𝐢−𝐍𝟐,𝐢+

𝐍𝟐

𝟏𝐍×

𝐢−𝐍𝟐

𝐢+𝐍𝟐 𝐱𝐢 − 𝐱 𝟐

Fractal

𝐇 =𝟏 + 𝟑

𝟒 × 𝟐,𝟑 + 𝟑

𝟒 × 𝟐,𝟑 − 𝟑

𝟒 × 𝟐,𝟏 − 𝟑

𝟒 × 𝟐

𝐆 =𝟑 − 𝟑

𝟒 × 𝟐,−𝟑 + 𝟑

𝟒 × 𝟐,𝟑 + 𝟑

𝟒 × 𝟐,−𝟏 − 𝟑

𝟒 × 𝟐𝐘𝐇 𝐢 = 𝐱𝟐𝐢 × 𝐆 𝟒 + 𝐱𝟐𝐢+𝟏 × 𝐆 𝟑 + 𝐱𝟐𝐢+𝟐 × 𝐆 𝟐 + 𝐱𝟐𝐢+𝟑 × 𝐆(𝟏)𝐘𝐋 𝐢 = 𝐱𝟐𝐢 × 𝐇 𝟒 + 𝐱𝟐𝐢+𝟏 × 𝐇 𝟑 + 𝐱𝟐𝐢+𝟐 ×𝐇 𝟐 + 𝐱𝟐𝐢+𝟑 × 𝐇 𝟏

𝐰𝐡𝐞𝐫𝐞 𝐢 = 𝟏 𝐭𝐨 𝐌/𝟐

𝐀𝟏𝟐𝐢 = 𝐦𝐢𝐧 𝐦𝐢𝐧 𝐘𝐇 𝐢 −𝐧

𝟐𝐭𝐨 𝐢 +

𝐧

𝟐,𝐦𝐢𝐧 𝐘𝐋 𝐢 −

𝐧

𝟐𝐭𝐨 𝐢 +

𝐧

𝟐

Wavelet

𝐄𝐢 = 𝐱𝐢𝟐 + 𝐱𝐢

𝟐

𝐀𝟏𝟒𝐢 =

𝟏𝐍 𝐣=𝐢𝐣=𝐢+𝐍−𝟏

𝐄𝐣

𝟏𝟓𝐍

𝐣=𝐢−𝟓𝐍𝐣=𝐢−𝟏

𝐄𝐣

ESM

𝐄𝐢𝟐 = 𝐱𝐢

𝟐 + (𝐱𝐢 − 𝐱𝐢−𝟏)𝟐×

𝐣=𝟏𝐢 𝐱𝐣

𝟐

𝐣=𝟏𝐢 (𝐱𝐣 − 𝐱𝐣−𝟏)

𝟐

𝐀𝟏𝟑𝐢 =𝐄𝐢𝟒 − 𝐄𝐢

𝟒

𝛔 𝐄𝐢𝟒

BKM

………….………….………….

FilterCF

ANN autopicker – Implementation

AIC algorithm fails …..Examples comparing ANN picks with AIC and hybrid STA/LTA picksANN picks validated with modified STA/LTA showing good match


property estimation



ANN picks for synthetic hydrofrac data under varying SNR

Observations

Speed: When compared with contemporary (LBNL) autopickers,the ANN autopicker takes similar run times. However attributeevaluation and manual training data selection can lead toadditional computational times.Applicability: The autopicker has been tested with twoindependent geothermal datasets as well as synthetic hydraulicfracturing data and it has shown good results including in testswith very noisy data.Accuracy: The results compare favorably when compared withtwo contemporary autopicking algorithms.

Robustness: The workflow has shown ability to operate under awide range of noise and waveform characteristics.


property estimation



Part 3

Novel MEQ based geomechanical property estimation

Inversion results

High resolution velocity models

Elastic property prediction


• Time differential b/w p and s phase used to findthe distance to event• Amplitude of the strongest wave used tomeasure the magnitude of the event

Theory – hypocentral inversion

hypocenter Recording station

Calculate residuals

Obtain least square inverse of residuals

Assume event location

Use crustal model to predict arrival time

Iterate forthreshold


property estimation



Theory - velocity inversion

Adapted from Um & Thurber, 1987

Pseudo-BendingApproximateRay - tracing

Adapted from Thurber et al., 1980

Eberhart-Philips, 1990; Zhao et al., 1992; Scott et al., 1994; Eberhart-Philips & Reyners, 1997; Graeber & Asch, 1999

ART, PB algorithms

Thurber, 1983; Eberhart-Philips, 1986ART + PB algorithms

Velocity & location perturbation to model

Relocate & repeat velocity inversion

ART locations as first estimate

Damped least square solution to residual eq.

Iterate forthreshold

Hypoinverse results

06/13/2013 PhD Defense 2406/13/2013 PhD Defense 24Introduction AutopickingGeomechanical

property estimation



Event count Spatiotemporal spread Event uncertainties

Hypoinverse results

Typical raypath coverage De Natale et al., 1996

Increase array size – improve ray path coverage as well as usable catalog size

Improved inversion algorithms – TomoDD, etc

Better understanding of regional geology including the fault systems to understand observed seismicity patterns

RECOMMENDATIONS

Increase array size – improve ray path coverage as well as usable catalog size

Improved inversion algorithms – TomoDD, etc

Better understanding of regional geology including the fault systems to understand observed seismicity patterns

RECOMMENDATIONS

Improved array design

1. Ray path focusing to prevent ill conditioned inverse problem based on source – receiver information.

S1

R1

R2

Rn

…

Pair vectors

Ray Matrix

Singularity using Sabatier’s data angle.

2. Focal mechanism from microseismic moment tensor with inversion stability based on condition number.

d1

d2

d6n

a11 a12 a16

a6n1 a6n2 a6n6

.

.

.

...

.

...

.

.

.

.

m1

m2

m6n

.

.=

Over-determined system requires generalized inverse

of A for match

Stable inversion matrix

Minimize condition no.

Solid angle (proxy for receiver area) useful. ↑ solid angle (Ω)↓ cond. no.

Reduces impact of increased noise conditions on the stability of matrix inversion.

T

T

Improved array design

QF = W1 × QF1 + W2 × QF2 (W1 = 1.0 & W2 = 0.0)

QF = W1 × QF1 + W2 × QF2 (W1 = 0.0 & W2 = 1.0)

SimulPS results

Sample VP and VS maps at 1 Km depth level after SimulPS run

06/13/2013 PhD Defense 2906/13/2013 PhD Defense 2906/13/2013 PhD Defense 29Introduction AutopickingGeomechanical

property estimation



Improved velocity modeling – COSGSIM

Better estimate velocity (primary) based on seismic derived impedance (secondary)

Seismic Data

VP & VS

COSGSIM

VP Realizations (Gaussian Domain)

Normal Score Transformed VP & VS

VS Realizations (Gaussian Domain)

Final VP Realizations Final VS Realizations

Secondary Variable

Primary Variable

Normal Score Transformation

Inverse Normal Score

Transformation

Impedance Maps

Microseismic Data

COSGSIM –inputs & results

PDF & CDF of normal score transformed VPPDF & CDF of normal score transformed VS

Property estimation

Lame’s ParametersElastic PropertiesStress EstimatesAnisotropic Weakness

Here μ & λ are Lame’s parameters. K, E & σ are the elastic moduli, VE & VK are the stress

estimates (extensional and hydrostatic), ΔN & ΔT are the normal and tangential weakness

estimates and finally, FE is the normalized fracture aperture expandability.

ρ is the density estimate, VP is the compressional while VS is the shear wave velocity in the

medium, e is the crack density in the medium, br, bmax and α are aperture measurements

obtained from laboratory tests using empirical rock classification index.

Mavko et al., 2003

Beer et al., 2009

Tokosoz et al., 1981

Hsu et al., 1993

Rutqvist et al., 2003

Fracture Aperture


property estimation




Property estimation

Lame’s Parameter - μLame’s Parameter - λBulk Modulus - KPoisson’s Ratio - σExtensional stress - VEHydrostatic stress - VKNormal weakness - ΔNTangential weakness - ΔTNormalized fracture aperture expandability - FE


property estimation



Rock physics for characterization

Fluid Saturation↓ Vs or ↑ Vp/Vs & σ

Effective pressure↑ Vp & K

Fractures↓ Vp & Vs

Fracture opening↑ VE & ↓ K

Lithification↑ Vp/Vs, μ & K or ↓ σ

Porosity↓ Vp/Vs & K

Pore pressure↓ Vp

Gas↓ Vp, Vs & K

Fracture density↑ ΔT

Martakis et al., 2006; Berryman et al., 2002; Berge et al., 2001; Boitnott, 2003, Downton et al., 2008

Observations

High resolution models: Major limitations plague the availableMEQ data for this study but we have utilized geostatisticalsimulation involving seismic derived impedance map as a meansto constrain and improve the resolution of inverted phasevelocities.Seismic derived density for property estimates: We havedeveloped a framework for utilizing seismic and log deriveddensity maps to integrate with the velocity estimates to calculateelastic properties for interpretation and use.Seismic derived FZI for crack density estimates: We have usedseismic derived FZI maps to estimate crack density (e) bynormalizing the FZI values and using it as an estimate for “e”required for weakness estimation.Framework for 4D: We have established the framework for 4Dproperty estimation by using segmented catalogs and inversion toobtain time lapse velocity maps and using baseline seismicderived impedance as a constraint for time lapse high resolutionvelocity maps and subsequent geomechanical modeling.Introduction Autopicking

Geomechanical property

estimation



Part 4

3D surface seismic data analysis

Background

Seismic/ log attributes

ANN derived discontinuity

FZI


Attributes

Different measurements derived from the original seismic

data to help reveal features, relationships or patterns in the

data useful in interpretation.

What are seismic attributes?

CurvatureFaults and lineaments

CoherenceFaults and fractures

Instantaneous

FrequencyFracture zone, bed

thickness, shale

content & edges

Absorption

Factor (Q)Absorption

characteristics

of beds

Relative AIPorosities, sequence

boundaries &

unconformities

EnergyStratigraphic

features and

geo-bodies

Seismic and log derived attributes

Q factor, Instantaneous frequency & coherencyLog derived impedance, density & porosity

Are these derived properties valid?


property estimation



Discontinuity mapping

Original seismic, Laplacian and Edge preserving filters

ANN derived discontinuity & confidence

Seismic derived FZI maps

FZI1 & FZI2 maps

Input maps

Training data extraction

Sample Well

Observations

Seismic derived log property estimates: We have used ANNbased property prediction workflows from well logs to map usableproperty volumes such as porosities and densities which we usefor characterization.ANN based discontinuity mapping: We have carried outdiscontinuity mapping and used it as a framework to understandthe observed production regimes at specific test wells.

ANN based FZI mapping: We have used ANN based mapping todevise new fracture zone identifiers from seismic data which useseither independent (seismic) or integrated (seismic + logs /seismic + logs + MEQ) approach to fracture identification.


property estimation



Part 5

Hybrid FZI attributes

ANN derived hybrid FZI

Interpretations


Hybrid (seismic + log + MEQ) FZI

TRAIN

TEST


property estimation



FZI3

FZI4

Mapped fracture attributes

FZI3FZI4FZIAkFi


property estimation



FZI4 projections (horizons of interest)

Part 6

Integrated analysis

Discontinuity maps

FZI maps

Edge maps

Stress maps

Weakness maps


Fractured intervals (VE, ∆T & FZI4)


Horizon 1 (Training)Horizon 2 (Training)Horizon 3 (Testing)


property estimation



Discontinuity integrated FZI4 maps

FZI4 integrated with discontinuity at 500m & 1000m depths

Connectivity using FZI4 & discontinuity


property estimation



VE, discontinuity gradient & edge maps

FZI4, stress gradient & edge maps

Concluding remarks

Seismic and microseismic data is independently processed and analyzed to

obtain useful reservoir property estimates (including geometrical, structural &

geomechanical attributes) which together form the basis for our novel

fracture characterization workflow.

We have demonstrated:

Improved phase detection for poor quality passive seismic datasets

High resolution velocity modeling with poor MEQ data quality (using seismic derived constraints).

Geomechanical property estimates for fracture zone identification

Implementation of newly defined hybrid FZI attributes

Integrated analysis and interpretation to better understand reservoir behavior and plan future

field development.

Integrated Reservoir Characterization for Unconventional ...€¦ · evaluation and manual training...

Documents

Transcript of Integrated Reservoir Characterization for Unconventional ...€¦ · evaluation and manual training...