Research On Fusion Method Based On Probabilistic Kernel Principal Component Analysis

The collected seismic data contains a large amount of underground geological information,and the seismic attributes extracted from the original data directly reflect the various characteristics of underground geology.With the continuous updating of computer technology,the continuous optimization of mathematical algorithms,the types of seismic attributes are also increasing,the ability to extract geological information from seismic data is greatly enhanced,and the prediction of reservoirs through seismic attributes has attracted more attention.However,most of the predictions of reservoirs using seismic attributes are predictive of reservoirs through a single attribute,which not only fails to achieve prediction accuracy,but also has multiple solutions.Therefore,multi-attribute analysis techniques emerge as the times require,using multiple seismic attributes for comprehensive analysis..The main method is seismic multi-attribute fusion technology based on multi-source information fusion.Among them,multi-information fusion technology based on principal component analysis plays an important role in pattern recognition and data compression and dimension reduction.It is one of the methods of K-L dimensionality reduction transformation.Through certain criteria,complementing features,removing redundant information,and generating new data are the main ideas of principal component analysis.Since new data retains a large amount of redundant information while retaining the original data features,the prediction accuracy is improved.Solve the multi-solution problem of single seismic attribute analysis.The main research of this paper studies the principle of PCA,and uses the advantage of PCA's dimensionality reduction to reduce the dimensionality of many extracted sensitive attributes to achieve attribute fusion.Moreover,in order to better extract the nonlinear features in the sensitive attributes,the kernel function idea is introduced and the KPCA model is established.At the same time,the Bayesian probability theory is introduced,and the PKPCA model is established based on the model of the kernel function.Based on the PKPCA seismic multi-attribute fusion technology,the geological conditions can be better analyzed and the accuracy of reservoir prediction can be improved.At the same time,the PCA-based PKPCA lithology identification model and the PCA-based FDA lithology identification model were studied.The lithology feature identification classification is extracted by using multiple lithological related parameters.Aiming at the huge problem of large-scale seismic attribute data after the introduction of kernel function,the optimization problem of large-scale dataset kernel matrix is analyzed.The autocorrelation matrix optimization based on block idea and the sample library reconstruction method based on iterative idea are studied.Reduce the complexity of kernel matrix calculation while ensuring data accuracy.The spatial distribution of the layered reefs and the development of the favorable areas of the reservoirs were analyzed.The extraction of sensitive attributes was predicted by PKPCA multi-attribute fusion technology,and the results were in good agreement with the actual situation.