油藏描述中的地震相分析新技术.pptVIP

This diagram gives a synopsis of the process contained in SeisFacies. Describe the workflows directly off the diagram. Classifications can be carried out on traces, maps or samples (blocks) with outputs being seismic facies maps, attribute facies maps, and seismic facies volumes respectively. Stratimagic uses the neural network method while SeisFacies uses PCA for reduction of noise and data redundancy as a filter prior to the hierarchical classification. This is a powerful combination! Hierarchical is a two step process: 1) It characterizes meaningful population subsets and defines a representative for each one, and 2) it assigns individuals to the appropriate subset based on Euclidean distance. In addition, we have enhanced the interpretability of the hierarchical classification results by using a proprietary class sorting algorithm for the classification subsets. PCA - Statistical technique for data reduction Identifies the main elongation directions Allows dropping of the least variable directions In this 3-dimensional example, we define the the principal axes of elongation of the data clouds. Data points are projected along the new principal axes, reducing the number of dimensions from 3 to 2. For multi attributes, i.e.15, Eigenvectors and Eigenvalues help determine the number of output components. When the Eigenvalue is less than 1, the component is no longer representative of the information contained in the cloud of points. PCA is a process which is used to reduce noise and redundancy of the input data before the facies classification. PCA generates principal components (3D volumes) by extracting the most significant information from all of the input seismic attributes. First, the principal axes are located along the main data trends in the multi-dimensional crossplot space. Then the data cloud is projected into the principal axes and PCA components are generated. Only the first few PCA components contain useful information. The rest are mostly aff