Seismic Data Clustering Analysis Algorithm Based On Multiphase Level Set Image Segmentation

Application of data mining technology for geological reservoir prediction in oil development is of great significance.Since most of the geological data does not con-tain supervised information,so clustering technology has become the preferred method of geological structure analysis.Each seismic attribute of a work area can be regarded as an image,and because the seismic data is derived from the geological space azimuth information:the data in the same region has a large similarity,and the difference be-tween the data in the different regions is large.According to the above two points,this paper presents a geostructure cluster analysis algorithm based on multi-phase level set image segmentation.This algorithm includes two steps:data preprocessing,data segmentation and clustering.The initialization process consists mainly of two parts,the first part is the seismic attribute data cleaning,the second part is the dimensionality reduction of the seismic attribute data.The purpose of data cleaning is mainly focus on removing invalid data and replacing with valid one to ensure the integrity of the regional data;in many cases,not all dimensions of high-dimensional data sets for the real impor-tance of the underlying information is valid,the purpose of dimension reduction is to simplify the data,making the data more appropriate.In order to achieve this,we need to map high-dimensional data to low-dimensional data and to ensure that some impor-tant attributes(such as locality,other geometric properties)are preserved,thus greatly reducing the data without invalid information and redundant information,which can accelerate and simplify the computation in segmentation procedure in turn.In the data segmentation and clustering,the application of image segmentation algorithm based on local gray-level clustering of multi-phase level proposed by Li[1]is extended to the seismic attribute data segmentation of seven target regions.And because the original seismic attribute data information is large,even after the data dimension reduction pro-cessing,the dimension of the data is still more than one dimensions,the image segmen-tation algorithm based on local gray-scale clustering is extended to multidimensional data segmentation,so that it can deal with high-dimensional non-uniform seismic data.As a result,each data point is classified into one of seven classes according to its signs in three level set function.Finally,the clustering validity of the clustering is evaluated.The geometric struc-ture clustering analysis algorithm based on the multi-phase level set image segmenta-tion and the k mean algorithm which gets good effect in the actually geological reser-voir prediction are compared by three relative indexes:for the I and SD index,the clustering analysis algorithm of geologic structure with multi-phase level image seg-mentation is superior to the k mean algorithm.For the Dunn index,due to the influence of singularity,the clustering analysis algorithm of geological structure based on multi-phase level image segmentation is not as good as k mean algorithm.In conclusion,the clustering analysis algorithm based on multi-phase level image segmentation is superior to k mean algorithm in clustering effect.