高分辨率非线性岩性物性联合反演方法和应用

In the prediction of complex reservoir with high heterogeneities in lithologic and petrophysical properties, because of inexact data (e.g., information-overlapping, information-incomplete, and noise-contaminated) and ambiguous physical relationship, inversion results suffer from non-uniqueness, instability and uncertainty. Thus, the reservoir prediction technologies based on the linear assumptions are unsuited for these complex areas. Based on the limitations of conventional technologies, the thesis conducts a series of researches on various kernel problems such as inversions from band-limited seismic data, inversion resolution, inversion stability, and ambiguous physical relationship. The thesis combines deterministic, statistical and nonlinear theories of geophysics, and integrates geological information, rock physics, well data and seismic data to predict lithologic and petrophysical parameters. The joint inversion technology is suited for the areas with complex depositional environment and complex rock-physical relationship. Combining nonlinear multistage Robinson seismic convolution model with unconventional Caianiello neural network, the thesis implements the unification of the deterministic and statistical inversion. Through Robinson seismic convolution model and nonlinear self-affine transform, the deterministic inversion is implemented by establishing a deterministic relationship between seismic impedance and seismic responses. So, this can ensure inversion reliability. Furthermore, through multistage seismic wavelet (MSW)/seismic inverse wavelet (MSIW) and Caianiello neural network, the statistical inversion is implemented by establishing a statistical relationship between seismic impedance and seismic responses. Thus, this can ensure the anti-noise ability. In this thesis, direct and indirect inversion modes are alternately used to estimate and revise the impedance value. Direct inversion result is used as the initial value of indirect inversion and finally high-resolution impedance profile is achieved by indirect inversion. This largely enhances inversion precision. In the thesis, a nonlinear rock physics convolution model is adopted to establish a relationship between impedance and porosity/clay-content. Through multistage decomposition and bidirectional edge wavelet detection, it can depict more complex rock physical relationship. Moreover, it uses the Caianiello neural network to implement the combination of deterministic inversion, statistical inversion and nonlinear theory. Last, by combined applications of direct inversion based on vertical edge detection wavelet and indirect inversion based on lateral edge detection wavelet, it implements the integrative application of geological information, well data and seismic impedance for estimation of high-resolution petrophysical parameters (porosity/clay-content). These inversion results can be used to reservoir prediction and characterization. Multi-well constrains and separate-frequency inversion modes are adopted in the thesis. The analyses of these sections of lithologic and petrophysical properties show that the low-frequency sections reflect the macro structure of the strata, while the middle/high-frequency sections reflect the detailed structure of the strata. Therefore, the high-resolution sections can be used to recognize the boundary of sand body and to predict the hydrocarbon zones.