Research On Automatic Recognition Of Seismic Facies By Kernel Clustering Algorithm

Seismic facies identification is divided into different seismic structural units according to the intrinsic structure of seismic data.Seismic facies recognition is a typical pattern recognition problem.Firstly,several features(attributes)are extracted from seismic data,and then the clustering algorithm is used to divide the seismic facies.This paper is denoted to developing research of the new kernel clustering algorithm to identify seismic facies.The main research work in this paper is as follows:(1)Research on seismic facies identification based on approximate kernel kmeans clusteringThe kernel kmeans clustering algorithm needs all the samples to calculate the kernel matrix,so the computational complexity of the algorithm and the requirement of the memory matrix are high.The kernel kmeans is not suitable for large scale data such as seismic data.Therefore,the approximate kernel kmeans clustering(AKK means clustering)algorithm is used to identify the seismic facies.Firstly a small number of random samples are selected in all seismic data,it will limit the clustering center in high dimensional space which is constructed by the sampling.It only needs to compute the partial kernel matrix.Then,the AKK means is applied to seismic facies identification.Experiments show that seismic facies identification results based on approximate kernel kmeans clustering not only can identify reflection structure unit quickly and accurately,and the recognition results better than the traditional K-means clustering,spectral clustering and Nystrom clustering.(2)Research on seismic facies identification based on multi-view weighted approximate kernel kmeans clusteringBased on the approximate kernel kmeans clustering and multi-view learning algorithm,this paper proposes a multi-view weighted approximate kernel kmeans clustering(MV-AKK means clustering)algorithm which is applied to seismic facies identification.This algorithm firstly calculates each attribute sampling kernel matrix by a small amount of sample from all seismic attribute data.Secondly,the weight of each attribute is automatically determined according to the iterative constraint conditions of each attribute sampling kernel matrix.Finally,the sampling kernel matrix is weighted and it is used for AKK means clustering.The MV-AKK means clustering algorithm is applied to the Netherlands North Sea.The result is compared with kmeans clustering,AKK means clustering,SOM clustering algorithm and the single attribute clustering result by AKK means,which shows better clustering effect of MV-AKK means algorithm.(3)Research on seismic facies identification based on semi-supervised kernel mean shift clusteringThe semi-supervised kernel mean shift clustering,which we call SKMS clustering,is applied to seismic facies identification,this method can effectively combine the advantages of semi-supervised learning and kernel mean shift.It is not necessary to set the number of clusters,and the accuracy of seismic phase identification is improved by introducing a priori information to guide the clustering process.The clustering of theoretical data shows that the algorithm has higher recognition accuracy for several structural units in seismic facies.Actual data clustering results show that the algorithm can get a reasonable number of seismic facies,and it divides the seismic phase structure hierarchy and can distinguish the tiny micro layer.