Research On The Application Of Seismic Multi-attributes Fusion Methods

The changes of reservoir physical properties, the reservoir saturated fluidcomposition and other related information are hidden in the seismic data. However,the seismic attributes are the geometry morphology, characteristics of kinematics,dynamics, statistical which are related to the seismic wave and are derived by themeans of mathematical transformation. We may reveal the reservoir information byanalyzing the seismic attributes and making a calibration to eliminate the distortion.However, because of the complexity of the underground geological conditions and toomany influence factors of seismic information, there is a big uncertainty or ambiguitywhen predict the reservoir. As a result, it is infeasible by using any single seismicattribute to accurately predict the reservoir, oil and gas, or to describe the reservoircharacteristic parameters, etc. Comprehensive analysis on seismic multi-attributeappears to be very necessary under this circumstance.When talking about the multiple attribute fusion technology, we must consider toavoid the redundant of the attribute numbers and the numerous and diverse of theinput samples affecting the computational efficiency and precision of reservoirprediction first. In response to this problem, this paper analyzed the physical meaningand geological meanings of the single seismic attribute firstly, and conducted theattribute fusion application research by the method of PCA(Principal ComponentAnalysis) and ICA (independent component analysis); then, according to thecorrelation between the well logging data interpretation and the seismic attributes ofthe seismic trace near well, we exchanged the seismic attributes to reservoir physicalproperties(porosity),than calculated to the cross-borehole by using multi-attributelinear regression method and the partial least-squares regression method.The main research work and the main results of this paper listed as follows:1. Studied the seismic multi-attribute optimization analysis based on the methodsof PCA and ICA. Compared to PCA, ICA becomes a powerful tool whendecomposing the independent information in the observation data. It not only uses thesecond order statistics characteristic of the signal, but also uses the higher-orderstatistics characteristic. We can optimize the most sensitive, independent seismicattributes by using it for high order statistical characteristic seismic attribute analysis.Using the method of properties fusion, we can better analyze the geological condition, enhances the precision of reservoir prediction.2. Combined the ICA and the seismic spectrum decomposition technique. Thispaper studied the application of broadband instantaneous spectral data fusion based onICA method. It used the spectral decomposition method based on short-time Fouriertransform to get a series of a total instantaneous frequency spectrum attribute datavolume, and proposed a fusion method of instantaneous spectral data slices which wasbased on ICA method. Through the actual data processing, it is proved that thismethod can not only reduce the dimension of the original data, put forward theindependent properties, but also can be applied to the seismic facies classificationeffectively.3. When using the seismic multi-attribute to predict the reservoir parameters(porosity), this paper used the multi-attribute linear regression and partial leastsquares regression and compared the effect of the two methods. The biggest limitationof the application of multivariate linear regression is the multiple correlations amongindependent variables. This variable multiple correlations would cause seriousdamage to parameter estimation and expand the model error, which can seriouslyaffect the effectiveness and reliability of reservoir parameter prediction. And by usingpartial least square regression modeling, it can realize regression modeling (multiplelinear regression analysis), simplify data structure (principal component analysis(PCA) and correlation analysis between the two groups of variables (canonicalcorrelation analysis) under this algorithm. In particularly, the partial least-squaresregression uses the method of decomposing and screening the data information. It caneffectively extract the comprehensive variables which are strongest to the explanatorysystem, and eliminate the multiple correlation information and meaninglessinterference. The study shows that it is feasible for applying the partial least-squareregression method to the seismic reservoir parameter prediction.