A method using Neural networks and directive attributes to detect and display fluid indicators in seismic cubes

Paul Meldahl Statoil ASA

Content

Introduction

Examples of detected fluid indicators

The detection method

Conclusions

From the Gym Hall to the Virtual Reality

“Funn” , Stig Kvendseth

3D seismic interpretation require transparency in the seismic cubes

( )

A section from a seismic cube

A section from ”TheChimneyCube”

How to increase the transparently of the seismic cubes?

Enhance, detect and display selected classes of objects at a time.

How to enhance, detect and display selected class of objects at a time ?

Use our object detection method founded on supervised neural network and directive attributes

Object- detection

3D interpretation require transparent seismic cubes

Chimneys and pock marks indicating focus areas of fluid flow above the top reservoir

Example of transparent seismic cubes:

A section from a seismic cube

The same section after detection of seismic chimneys

Object- detection

Fluid indicators may be diffuse and difficult to distinguish

How to enhance and detect diffuse objects in the seismic?

Enhance, detect and display selected classes of objects at a time.

How to enhance, detect and display selected class of objects at a time ?

Use our object detection method founded on supervised neural network and directive attributes

Example of transparent seismic cubes:

Salt, seismic chimneys , reservoir, shallow reflectors and pock pock marks

Object detection

Example of a transparent seismic cube; seismic chimneys on top of reservoir

Example of transparent seismic cube;Seismic chimneys and faults indicate where the faults may leak

Directivity principle in the detection method :

The size and orientation of windows to extract a set of attributes follow the objects we wish to detect

x

Verticalfor chimney’s

Horizontal for ‘DHI’ s

Dippingfor faults

Dip-steered for flexible bodies and interfaces

xx

The detection method: Overview

The detection method: Workflow

• Read the seismic cube from the conventional seismic interpretation system• Choose one class of objects to be detected and the associated set of seismic attributes• Pick training points in the seismic cube on objects to be and not to be detected • Start the automated process of training neural network• Start the automated process of detection • Display the transparent cube where the selected class of objects are enhanced• Write the transparent cube back to the conventional interpretation system

Prediction of fluid flow by detection of Chimney and Faults may have impact in Exploration

Transparent seismic cubes show spatial relations between seismic objects A powerful and flexible method for the Detection of Seismic Objects as: Chimneys, Faults, 4D objects, Reflectors, Salt and DHI’s Is presented

The detection is efficient because it’s user friendly and uses the combinations of directive attributes and supervised neural network

A challenge is to develop semi automated mapping procedures which take advantages of transparency

Conclusions

From real sniffing to “seismic sniffing” by

neural networks and directive seismic attributes ?

From ‘The Prize’ by Daniel Yergin

Acknowledgement

To Statoil and dGBfor making object detection

a reality