Seismic facies analysis from pre-stack data using self-organizing maps

Abstract

In facies analysis, seismic data are clustered in different groups. Each group represents subsurface points with similar physical properties. Different groups can be related to differences in lithology, physical properties of rocks and fluid changes in the rocks. The supervised and unsupervised data clustering are known as two types of clustering architecture. In supervised clustering, the number of clusters is predefined, while in unsupervised clustering, a collection of patterns partitions into groups without predefined clusters. In this study, the pre-stack data clustering is used for seismic facies analysis. In this way, a horizon was selected from pre-stack data, followed by sorting of data using offset. A trace associated with each CDP is constructed, for which the first and second samples are related to the first and second offsets, respectively. The created trace is called consolidated trace which is characteristic of subsurface points. These consolidated traces are clustered by using self-organizing maps (SOM). In proposed pre-stack seismic data clustering, points with similar physical properties are placed in one cluster. Seismic data associated with hydrocarbon reservoirs have very different characteristics that are easily recognized. The efficiency of the proposed method was tested on both synthetic and real seismic data. The results showed that the algorithm improves the data classification and the points of different properties are noticeable in final maps.