--> --> Statistical Variable Salt Velocity Calculation by Neural Network Classification in the Central Gulf of Mexico
[First Hit]

AAPG/SEG International Conference & Exhibition

Datapages, Inc.Print this page

Statistical Variable Salt Previous HitVelocityNext Hit Calculation by Neural Network Classification in the Central Gulf of Mexico

Abstract

Abstract

Due to salt tectonics, salt Previous HitvelocityNext Hit in the Gulf of Mexico (GoM) may vary due to sediment inclusions and sutures. The maximum slowdown may be up to 20% of the clean salt Previous HitvelocityNext Hit according to well log information. Therefore, using a variable salt Previous HitvelocityNext Hit to build a more accurate Previous HitvelocityNext Hit model is critical to improve base salt interpretation, subsalt imaging and well ties. Amplitude-based methods have been widely used to derive variable salt Previous HitvelocityNext Hit in industry by root mean square (RMS) amplitude, envelope, or others, mostly single seismic attribute driven with a constant scalar between the attribute and salt Previous HitvelocityNext Hit.

A neural network is an adaptive system that changes its structure based on external (supervised) or internal (unsupervised) information during the learning phase. It is trained to classify the input data into a given number of classes. According to different seismic attributes (for example, envelope, RMS amplitude, absolute amplitude, and gradient, and others), the neural network classifies the salt image into a given number of classes. The different classes are scaled to different salt velocities with different scalars, which are decided according to the image tie to the well.

This statistical neural network method was applied to a large reprocessing project in the US central of GoM covering 55000 km2. The variable salt Previous HitvelocityNext Hit was calculated based on a constant salt Previous HitvelocityNext Hit reverse time migration flood. With several seismic attributes of the image, unsupervised classification was run to classify the image inside the salt into certain classes. According to the well logs in this survey that penetrate through the salt, an initial scalar was assigned to each class to derive variable salt velocities. A quick demigration with clean salt Previous HitvelocityNext Hit and remigration with the variable salt Previous HitvelocityNext Hit was done in some local areas to check the base salt marker tie, and then the Previous HitvelocityNext Hit scalars were adjusted accordingly. Localized RTM migration was followed along certain well locations to check the base salt tie for further Previous HitvelocityNext Hit scalar adjustment. The neural network method is much easier and faster to adjust the salt Previous HitvelocityNext Hit scalars for a more accurate salt Previous HitvelocityTop and base salt marker tie, which leads to more accurate base salt interpretation and enhanced subsalt imaging.