Research And Application Of Anomaly Data Detection And Repair For Seismic Facies Recognition

Data mining has high requirements for data quality.Abnormal data in data has a negative impact on the efficiency and effect of data mining.It is very important to repair or exclude them.Accurate abnormal data detection and repair plays a key role.In the field of seismic facies classification,seismic facies classification plays an important role in stratigraphic environment analysis and reservoir prediction.However,the anomaly data in seismic data have a certain impact on the quality of seismic facies classification.Repair or elimination of abnormal data can improve the quality of seismic data.On the basis of introducing the decision system,the sample set are divided into nearest neighbor sets,reverse nearest neighbor sets and shared nearest neighbor sets according to the sample decision attributes,which construct the local expansion domain.Then the Gaussian kernel is used to calculate the density of local extension domain,and the correlation weights are formed by combining the pairwise constraints of the samples with the density.Finally,the density anomaly factor Semi-supervised Local Density based Outlier Factor(SLDOF)of each sample is defined to evaluate the degree of sample anomaly.The detection of abnormal samples is realized by the evaluation of each sample anomaly.Through the experiments with simulation data set and real data set,it is confirmed that SLDOF can be applied to abnormal detection of whole-labeled or fractionlabeled samples,and performance is fairly effective.To adress the probblem of lacking of effective information for deletion and repair of abnormal sample,a sample exception repair algorithm based on correlation analysis Correlation analysis Based Data Repair(CBDR)is proposed.Firstly,the conditional attribute correlation graph is obtained by density clustering using sample density information.Then,the correlation analysis was carried out to divide the conditional attribute correlation clique,and the repair cost were calculated.Finally,the target sample is selected according to the minimum cost principle,and partial local data repair is carried out according to the relevant conditional attribute groups.Through the experiments with simulation data set and real data set,it is confirmed that CBDR is capable of effective repair of abnormal data.To address the problem of abnormal data in seismic data,SLDOF and CBDR are applied to seismic dataset.The anomaly detection and anomaly repair are performed on them respectively.Then the processed seismic data are classified and predicted.The results of experiments show that both SLDOF and CBDR are effective for seismic data.