Seismic Facies Analysis Based on Spectral Clustering with Waveform Characteristic Vector

Abstract

Based on the principle of seismic sedimentology, the feature vectors of seismic waveforms are extracted along stratum slices, and spectral clustering analysis is introduced to classify seismic facies. Spectral clustering is an unsupervised machine learning algorithm. Its essence is to simplify the expression of high-dimensional seismic data in the form of feature vectors, which belongs to the process of dimensionality reduction. Considering the traces with specific time windows in the seismic work area as nodes of the graph and the similarity between traces as the weight of the edges, a graph model can be constructed. Spectral clustering must determine the best segmentation method to complete the segmentation of the graph, so that different types of sedimentary characteristics can be distinguished. Physical model and actual data processing and analysis demonstrate that this method is capable of dividing sedimentary facies characteristics and is a new kind of facies analysis tool for reservoir classification, which has good application prospects.

Keywords

• machine learning
• spectral clustering
• waveform clustering
• seismic facing
• seismic sedimentology