M8-2_Train the Estimation Model

Train the Estimation Model

Petrel Geophysics

Advanced Interpretation (2)

Objectives

Train the estimation model.

– Use the process with neural net as method.

Genetic inversion.

– Perform Well log conditioning.

– Choose genetic inversion as an attribute.

– Define correctly parameters.

Train Estimation Model: Neural Nets (1)

1. Generate one or more Attribute cubes from the same

seismic volume.

2. Open the Train Estimation Model process under Utilities.

5. Set a selection of

Training data.

6. Click Correlation analysis to calculate the correlation coefficients

between the input volumes.

4. Drop in the mother seismic volume in the

Seismic drop-in area.

3. Use Create new and Classification options. From the drop-down

menu under Data type, choose Seismic.

Train Estimation Model: Neural Nets (2)

7. Click the Settings tab and specify the number of classes under Unsupervised.

8. Set the Training parameters or use the defaults.

9. Display the Time slice from the newly created

Classified volume.

Classe

s

0

1

2

Genetic Inversion required inputs are limited to the seismic amplitude cube and the

Property well logs

Introduction to Genetic Inversion

Benefits of Genetic inversion

• Lithology identification.

• Facilitate interaction between seismic interpreters with geologists,

petrophysicists, and reservoir engineers.

• Inverted data set can be used directly to constrain initial reservoir models.

• Improve volumetric estimations.

• Target wells more precisely.

A new approach to derive any petro-physical property linked to the seismic amplitude

volume using a non-linear seismic inversion, multi-layer neural network as well as a

genetic algorithm is to combine them to provide a powerful and straightforward

seismic inversion.

Genetic Inversion: Overview

Genetic inversion

Well log

Low frequency filter True amplitude

Seismic volume

Property cube

Genetic Inversion: Well Log Conditioning

At a minimum, well logs must be

despiked and low frequency

filtered to match the seismic

resolution to show a deflect

where a seismic reflector displays.

• Porosity (black left)‏

• Porosity filtered (red middle)‏

• Seismic (black wiggle right)

Genetic Inversion

2. From Stratigraphic

methods, choose

Genetic inversion.

3. Under the Input/Output tab, drop in the Input seismic volume and define the output name.

(Leave it unselected and it will be given a default name.)

1. Right-click the seismic

volume in the Input

pane and select

Volume attributes.

Genetic Inversion: Parameters

4. Go to the Parameters tab, which is divided into three sections:

A) Learning input: Define the input for the training process.

B) Settings: Adjust the 3D operator for the Neural nets.

C) Advanced options: Set the quality criteria for stopping the

training process. The default values are set to fit most data sets.

A

B

C

Acoustic impedance cube

EXERCISE

p417-429