AVO inversion based on inverse operator estimation

Yin Xing-Yao,Deng Wei,Zong Zhao-Yun

DOI: https://doi.org/10.6038/cjg20160426

2016-01-01

Abstract:Amplitude variation with amplitude or angle (AVO/AVA) inversion has been widely utilized in exploration geophysics to estimate the formation of elastic parameters underground. The traditional inversion algorithms are mainly optimization, while the AVO inversion based on inverse operator estimation is to inverse directly. The key is to find the subspaces which exist inverse mapping instead of searching for the solution directly as optimization algorithms do. Aki&Richards approximation is utilized to establish the inversion objective function. L1 norm constraint is considered on the basis of reasonable initial model in order to improve effciency and stability during AVO inversion process. Inverse operator estimation algorithm utilizes numerical approximation of objective function in empirically constrained subspaces. The existence of inverse mapping inside these subspaces is supposed, and thus the existence of its numerical approximation is also supposed. This numerical approximation is used for the prediction of the solution. It can be divided into the following four steps: (1) initialization, (2) selection, (3) prediction, (4) solution space correction. Since this method is approximate, the algorithm is arranged into iterative cycles and the solution is gained successively. Model test in which noise immunity and stability based on L1 norm constraint are considered lays the foundation for the actual data inversion. In the model test, the inversion results based on synthetic seismogram with different SNR noise remain good consistency with the model value. L1 constraint is necessary, a relatively stable result can he obtained in multiple inversion processes and it helps reduce the number of singular values. The seismic data is from China's land, inversion quality can be evaluated by well data and seismic data. Inversion results of actual seismic data maintain good consistency with logging curves. Reliability of the solution, noise immunity and convergence speed can be improved by increasing the constraint requirements and considering reasonable initial models. Poor lateral continuity when dealing with actual data can be reduced by optimizing the objective function and the initial model. In general, AVO inversion based on inverse operator estimation is effective and potential.