EEI ATTRIBUTES FOR FLUID DISCRIMINATION USING FUZZY LABELED MULTICLASS SUPPORT VECTOR MACHINE

摘要

In exploration seismology, automatic seismic facies analysis to discriminatedifferent facies and fluid content is an essential task to reduce future drilling risks. Thereare different seismic attributes as learning features and various learning methods forautomatic seismic facies analysis. Previous studies have proved that selecting efficientseismic attributes is more crucial than the learning method. Therefore, it is logical to paymore attention to the choice of proper attributes. The extended elastic impedance (EEI)attributes belong to prestack seismic attributes, and they are functions of compressionalvelocity, shear velocity, density, and chi angles. The Chi angle is the virtual incidentangle and changes between -90 to +90 degrees.The innovative method demonstrates the role of fluid replacement modeling(FRM) for the supervised selection of EEI attributes at suitable chi angles as inputfeatures to train an intelligent model for the discrimination of reservoir fluid contents.The method starts with FRM to model different fluid contents of the reservoir(100 brine, 100 oil, and 100 gas) using borehole data. Then, efficient EEI (Chi)logs are selected according to the results of the EEI template analysis. Thus, EEI seismicattributes at selected Chi angle are calculated from prestack seismic data by amplitudeversus offset (AVO) analysis and EEI inversion. Then, labeling of the EEI attributes isperformed by fuzzy c-mean clustering (FCM). By considering membership functions, afuzzy concept is an appropriate tool for soft clustering and an appealing method forseismic interpretation. Afterward, a classifier model of the multiclass support vectormachine (SVM) is trained using the fuzzy labeled samples to predict the fluid type ofunseen data.The method was applied to a 3D prestack seismic data of an oil sand reservoir inthe Persian Gulf to predict the fluid distribution map at the top of the reservoir. Thereservoir contains a considerable amount of gas cap. Only one borehole data drilled in theoil column is available for FRM and fluid EEI template analysis. The available fluiddistribution map confirms the accuracy of the resulted fluid distribution map based on themodeling of all the wells in different locations of the reservoir. This confirmation provesthe application of the proposed method in fluid pore identification.