A better Horizon Auto Tracker - Powered by Machine Learning
Abstract
Computer-assisted horizon interpretation on 3D seismic data has been commercially available for almost 15 years. The process is generally referred to as 3D horizon autotracking and is a powerful method when interpreting seismic horizons. The desire of a new autonomous horizon tracker is growing in the market. This thesis is combining the fields of seismic interpretation and data science in order to perform the early testing on a software aiming to fill this gap in the market. The goal is to find out if the new algorithm based on machine learning is a good replacement for the 3D horizon autotracker regarding a more effective and time-saving workflow. The technology supports radial basis functions (RBF) that enables the IntelliTracker to store information acquired during the workflow. This is necessary because of the complex nature of the geology and the varying quality of seismic data. The IntelliTracker was tested on the NH0301 dataset covering the top reservoir unconformity in the Troll field. Testing was performed by interpreting the top reservoir horizon with the existing 3D autotracker and then tracking the same horizon with the new method. The results from the testing prove that the current autotracker is still leading on several areas. The IntelliTracker can, however, be used as a guidance tool by the interpreter for a more effective workflow. Although the ML technologies have developed rapidly over the last couples of years, the interpretation done by a machine will always need to be quality checked.