Study On Prediction Method Of TOC Content And Brittleness Index Of Shale Reservoir

The prediction of "sweet spot" of shale gas is the key to the exploration of marine shale gas in Sichuan Basin.The purpose is to find the area where natural gas is rich and easy to form fracture network by hydraulic fracturing.The enrichment degree of natural gas can be evaluated by TOC content and gas content,which are positively correlated.The difficulty of hydraulic fracturing in reservoir rocks can be evaluated by the brittleness index.The shale with higher brittleness index is more likely to fracture after fracturing.At present,in shale gas exploration,both the indirect calculation of brittleness index and the multivariate fitting calculation of TOC content are mainly based on the three elastic parameters of prestack AVO inversion results.On the basis of previous studies,improved prediction methods are proposed to overcome the defects of conventional methods for predicting TOC content and brittleness index,such as insufficient fitting accuracy,accumulated error caused by indirect calculation and so on.Firstly,taking Dingshan area in Southeast Sichuan as an example,the shale wedge model and the actual two-dimensional geological model are designed.Through acoustic wave equation forward modeling,the influence of formation thickness,gas content and dominant frequency of wavelet on seismic response characteristics of shale is analyzed quantitatively.Then,based on the analysis of sensitive parameters and AVO forward modeling of shale reservoir,the prestack AVO inversion is performance in this area.Aiming at the TOC content prediction of shale reservoir,a new TOC content prediction method of deep learning is proposed in this paper.Through the intersection analysis of regression fitting method,the density and P-S wave impedance are selected which has the highest correlation coefficient with the measured TOC curve.Based on the classical deep feedforward neural network model,the nonlinear network model of TOC prediction is obtained by training the nonlinear relationship between the above three parameters and TOC.Then the inversion results of three parameters obtained from prestack AVO inversion are input into the network model to comprehensively predict the TOC data volume.In the actual application of logging and seismic data,the accuracy of TOC prediction result by this deep learning is better than that by regression fitting prediction,which effectively avoids the shortcomings of the regression fitting prediction by using a single parameter for linear regression fitting.It is proved that the prediction method of deep learning is realizable and practical.Aiming at the prediction of shale brittleness index,a novel direct inversion method based on NSGA-II algorithm is proposed in this paper.The forward operator BI＿Zoeppritz equation of the new method is the transformation form of the exact Zoeppritz equation about the brittleness index,P-wave velocity and S-wave velocity.There is no assumption and no loss of accuracy in the derivation process.This equation avoids the accumulated error due to the indirect conversion of elastic parameters.A multi-objective function is established to control the quality of invertion,which can not only reduce the error of RMS,but also improve the correlation degree.Then the fast non-dominated sorting genetic algorithm(NSGA-II)is selected to obtain the global optimal solution for the above-mentioned double objective function.For the purpose of reducing the calculation time and constraining the whole inversion process,an initial low frequency model is established according to the actual data,meanwhile,an optimized search window is defined and applied.The results of the experiments of several models prove that the method is suitable and antinoise.The brittleness index is successfully predicted by applying the new method to practical data,and it is consistent with the actual logging curve,which shows that it is feasible to use this method to directly invert the brittleness index from the actual seismic data.