Reservoir Prediction Technology-Based Kernel Method Of Well Logs And Seismic Multi-attribute In Clastic Rock

With the developmernt of new theory and tecthnology prentation and itsapplication, explorationist pay attention to the unconventional tight clastic reservoirbacause of its industty latent capacity and economic worth,and a series of mothod andtechnology is presented.With the complication and peculiarity of tight clasticreservoir,which is caused by lower porosityand lower permeability,the accuracy can’tbe figured out in reservoir prediction. Except application of advanced technology anddevice，the multi-disciplinary techniques is need，which combines geology, welllogging, seismic exploration, nonlinear, multi-disciplinary techniques.In this paper, the first part summarily review the kernel methods and principlesof statistical learning theory,and introduces the support vector machine classifier,support vector machine regression, kernel Fisher discriminant method and kernelprincipal component analysis method.Moreover,the capabilities and prospects ofapplication of these methods is studied in the prediction of reservoir parameters, the Pand S data forecast, identification of lithology, fluid identification andmulti-dimensional seismic attribute optimization analysis,then experimental resultswere analyzed and discussed. Finally, with the optimized seismic attributes, combinedwith the waveform clustering spectrum analysis and acoustic impedance inversionresults, the favorable of oil and gas area is presented. The main achievements andinnovation are as follows:（1） Propose least squares support vector regression machine for shear waveforecast. On the basis of depth study of support vector machines, shear waveprediction of empirical formula and least-squares fitting method, least squares supportvector machine based on statistical learning theory is used for fitting the measuredS-wave data,and obtains a better prediction accuracy and adaptive ability. In thecalculated shear wave data,it gains a variety of elastic parameters. (2) Kernel Fisher discriminant method is introduced to lithologic identificationin tight clastic rock, and to explore the potential and prospects of its application inthis field.For lithology identification difficulties of XC2member， Fisherdiscriminant analysis and kernel Fisher Discriminant Analysis are used to calculatethe elastic parameters with well logs,elastic parameters and combinations.The resultsshow that kernel Fisher Discriminant Analysis can well identify the sandstone andsiltstone.(3)According to the nonlinear characteristics of fluid idenfification in the tightclastic reservoirs,KFDA and LS-SVM based on principles of statistical learning wasput forward.Firs take XC2member for example,kernel Fisher DiscriminantAnalysis,successfully applied to lithologic identification, is introduced into reservoirfluid identification.The results show that method can better distingiush between gasreservoir and gas with the water layer.Secondly,summarizes the existing1-against-1,1-against-rest and least squares support vector classifier, least squares support vectormachine was put forward for fluid identification,which improve the classificationaccuracy of the data.(4)Four common kernel function of KFDA and SVM were analyzed andcompared based on building technology system of lithology identification and fluididentification.The result shows support vector machine training and classificationspeed is superior to neural network,the classification accuracy of the Gaussian kernelfunction.(5) Throgh the analysis and compare of seismic attributes optimzation method inparticular PCA,KPCA algorithm is in-depth studied,then simulation is obtained.Whenfaced with the large-scale data set, Matrix-based Kernel Principal ComponentAnalysis is used for seismic attribtes optimation,the result suggest that the optimatedattributes effectively characterise of the reservoir.(6) In this article,pectrum decomposing,waveform clustering,seismic inversionmethods is applied to the prediction of sandstone district area; Combination KPCAattributes with the result of waveform clustering and seismic inversion,folding graphof favorite area shall be provided for range of profitability for tight sand of XC2member,which shows good prospects for exploration and development.