Workflow for Extraction of Seismic Data From Multiple Synthetic Seismic Models by Using Clustering Algorithms

Abstract

Seismic data can be interpreted by two main methods; manual interpretation by geoscientist guided by algorithms, or by interpretation of seismic bodies using automatic algorithms. We have examined the second approach, because when interpretation is done using algorithms, the interpretations are comparable. This requires a workflow for extraction of seismic data from multiple synthetic seismic models. Different seismic attributes sets are calculated for these models. For selecting the best performed seismic attributes sets for the geological feature, a literature database is used. Then the seismic attributes are clustered using alternative clustering algorithms. Therefore 4 groups of clustering algorithms are tested: The first group is the Partitional Clustering Group, the second the Hierarchical Clustering Group, the third the Density-Based Clustering Group and the forth the Probabilistic Clustering Group. Each group consists of a variety of different algorithms. To achieve good clustering results, the data must be correctly selected and conditioned, and the appropriate seismic attributes and clustering algorithms must be determined. These decisions are discussed in this study and it can be concluded that it is important to know what the interpreter expects from the seismic data in order to select appropriate seismic attributes. A second important issue concerning seismic attributes is the appropriate scale, which should reflect the size of structures to be resolved. It is important to choose attributes that express the same features but with different values. It is also important to subdivide large volumes of seismic data to reduce the number of different clusters, particularly when handling real seismic cubes. For clustering, additional experimentation is necessary to estimate the number of clusters required and to choose the most effective clustering algorithm. To find the best clustering algorithm, it may be important to anticipate the shapes of resulting clusters. To determine this, further tests are necessary and additional techniques to determine these automatically are recommended. The automatic testing environment has been applied to a real seismic volume using the New Zealand data pack. The experience gained from the workflow on the synthetic seismic cubes was used to guide this work.

AAPG Datapages/Search and Discovery Article #90332 © 2018 AAPG International Conference and Exhibition, Cape Town, South Africa, November 4-11, 2018