Application of Manifold Recognition Target Identification Method in Seismic Exploration

Currently, there are more than 200 types of seismic attributes extracted from seismic data, the high-dimensional features of the data have become increasingly obvious. Although the increasing number of seismic attribute parameters is beneficial for researchers to understand seismic data, the massive amount of data also leads to redundancy and makes it difficult to further explore deeper buried information in seismic attributes, thereby reducing the accuracy of reservoir prediction. Manifold learning projects high-dimensional data into low-dimensional space by maintaining the local structure of the data, and mines and discovers the inherent characteristics and regularities hidden in the data. It is a new field of seismic attribute optimization research. This study is based on the application of manifold learning algorithms from cognitive science technology, comparing the advantages and disadvantages of the Isometric Mapping (ISOMAP) and Multidimensional Scaling (MDS) for extracting seismic attribute features, reducing the dimensionality of seismic attributes and optimizing the attributes. Both theoretical model analysis and practical application show that manifold learning has better clustering analysis ability and feature extraction performance in dealing with nonlinear problems, Seismic attributes extracted by ISOMAP are more accurate than those extracted by MDS in characterizing the distribution characteristics of favorable reservoirs, which provide powerful tools for subsequent reservoir characterization, sweet spot identification, and seismic interpretation.