Unsupervised seismic facies analysis using sparse representation spectral clustering*
Wang Yao-Jun, Wang Liang-Ji, Li Kun-Hong, Liu Yu, Luo Xian-Zhe, and Xing Kai
1. School of Resource and Environment, University of Electronic Science and Technology of China, Chengdu 611731, China.
2. Qiushi College, Taiyuan University of Technology, Taiyuan 030600, China.
Abstract Traditional unsupervised seismic facies analysis techniques need to assume that seismic data obey mixed Gaussian distribution. However, fi eld seismic data may not meet this condition, thereby leading to wrong classifi cation in the application of this technology. This paper introduces a spectral clustering technique for unsupervised seismic facies analysis. This algorithm is based on on the idea of a graph to cluster the data. Its kem is that seismic data are regarded as points in space, points can be connected with the edge and construct to graphs. When the graphs are divided, the weights of the edges between the different subgraphs are as low as possible, whereas the weights of the inner edges of the subgraph should be as high as possible. That has high computational complexity and entails large memory consumption for spectral clustering algorithm. To solve the problem this paper introduces the idea of sparse representation into spectral clustering. Through the selection of a small number of local sparse representation points, the spectral clustering matrix of all sample points is approximately represented to reduce the cost of spectral clustering operation. Verifi cation of physical model and fi eld data shows that the proposed approach can obtain more accurate seismic facies classification results without considering the data meet any hypothesis. The computing efficiency of this new method is better than that of the conventional spectral clustering method, thereby meeting the application needs of fi eld seismic data.
This work was supported by National Natural Science Foundation of China (Nos. U1562218, 41604107, and 41804126).
Corresponding Authors: Wang Yao-Jun (Email: yaojun.wang@uestc.edu.cn)
E-mail: yaojun.wang@uestc.edu.cn
About author: Wang Yao-Jun, associate professor, graduated from Anhui University in 2009 with a bachelor’s degree in Mathematics and Applied Mathematics. In 2015, he obtained a Ph.D. in Geological Resources and Geological Engineering from the China University of Petroleum (Beijing). He was a lecturer at the University of Electronic Science and Technology from 2015 to 2019 and an associate professor from 2019 to present. His current research interest is artificial intelligence in geoscience and geophysical inversion.
2020, Vol. 17 
