| This paper mainly studied the application of Support Vector Machines(SVM) method on reservoir prediction. The main work of the paper is summarized as follows:(1) This paper began with an analyzation of the classification of seismic attributes in reservoir geophysics, extracting methods and the basic work process of seismic attributes analyzing technology.(2) A hydrocarbon bearing and aquiferous model was designed, and with wave-equation forward, we gained the synthetic seismogram. We extracted 18 kinds of seismic attributes including Root Mean Square Amplitude, Instantaneous Frequency and Instantaneous Phase. Then employed range standardization method, the principal component analysis and correlation analysis to accomplish the standardization, characteristic dimension reduction, the sensibility of attributes value analyzation, achieved 8 kinds of seismic attributes such as Instantaneous Phase, which are very sensitive to the given target layer.(3) Correlative algorithm of SVM can attain the global optimal solution, so it can overcome the defects that the appearance of over-fitting and under-fitting in the artificial neural networks. According to numeric experiment, the SVM machine learning model has a better capability of robustness and generalization than BP, RBF network learning model. Applying it on the model's aquosity, oil-gas-bearing possibility prediction, the input parameter is the optimized seismic sensitive attributes, the output is defined as whether the given target layer is a reservoir. After using girding search method, we obtained the optimized parameter and formed the training model. the output of SVM classifier was almost identical with the theoretical model's; In the trail of wedge geological model's velocity and density, the predicted velocity has a relevance of 0.9877 with the previous one and its root mean square error is 0.0041; the relevance of density is 0.9885 while root mean square error is 0.004. The experiment shows our method is feasible.
|
PDF Full Text Request