Prediction of Geological Characteristic Using Gaussian Mixture Model

The multi-attribute classification has been widely applied in recent years as one of the direct approachs of seismic reservoir prediction. However, the use of multi-attribute seismic volume classification is a difficult and time-consuming process. In this paper, we propose a simple effective unsupervised approach based on the Gaussian Mixture Model (GMM) to identify special geologic characteristics (geobodies) including fractures, submarine channels, and reefs. The method proposed is a nonlinear statistical process of parameter learning, in which each kind of geobody is represented with one or multiple Gaussian distributions and given a unique cluster label. This method enables different geobodies to be directly extracted from 3-D seismic data. Experimental results on the real field data from Western China show that our method is effective in automatically detecting special geobodies and reducing the uncertainties in reservoir characterization prediction.