Three-dimensional full waveform inversion energy weighted gradient preprocessing method

Abstract

The invention relates a three-dimensional full waveform inversion energy weighted gradient preprocessing method. The method comprises the steps of (1) carrying out time second-order space finite difference method 3D forward simulation, storing a boundary wave field, and storing the maximum amplitude of reaching each speed grid point at the same time, (2) carrying out source wavefield reconstruction and residual wave field backward spreading at the same time, and carrying out corresponding time correlation to obtain a gradient, (3) reading the information stored in the step (1) and obtaining an energy weighted operator to carry out preprocessing of the gradient, and (4) obtaining the gradient operator of an appropriate step length action after preprocessing, and obtaining speed update quantity iterative update speed. According to the method, in the premise of not increasing computing amount, the problem of gradient operator deep and shallow layer energy imbalance caused by spherical wave spreading geometric diffusion and the like is solved, and the precision of deep layer speed imaging by full waveform inversion is improved effectively.

Description

technical field [0001] The invention relates to the technical field of geophysical exploration, and is mainly used for improving the accuracy of three-dimensional full waveform inversion of the longitudinal wave velocity field. Background technique [0002] Velocity is an important parameter to describe the condition of the subsurface medium, and the focus of geophysical exploration is how to recover the velocity information of all scales of the subsurface medium. There are many methods of speed analysis, but the desired effect is not always achieved. Tomography methods such as ray tomography and wave equation tomography can recover velocity information at low wavenumbers, and migration methods can provide reflectivity information at high wavenumbers. However, these methods cannot simultaneously recover velocity parameters at all wavenumbers. Full waveform inversion utilizes the full information of prestack seismic data to provide high-resolution imaging of subsurface media...

AI Technical Summary

Problems solved by technology

The main difficulties in 3D full waveform inversion are as follows: (1) Huge amount of calculation: 3D full waveform inversion is based on the iterative inversion method of 3D wave equation forward modeling. In one iteration process, it is necessary to calculate the wave equation of multiple shots Forward modeling and backpropagation of the error wave field, the two parts need to be differentially calculated at each time step, accompanied by a huge amount of calculation

(2) The problem of massive storage: the full waveform inversion method based on the adjoint state gradient operator requires to store the entire forward wave field, and the storage capacity in the three-dimensional case is intolerable. It can be said that most of the time is consumed in the I / O read and write

A large amount of calculation can be accelerated through a variety of parallel algorithms, while the I / O read and write speed can only be improved by improving hardware performance

(3) Calculation efficiency and accuracy of the gradient operator: the conventional full waveform inversion gradient operator is obtained by cross-correlating the time second-order partial derivative of the forward wave field of the source with the back wave field of the residual wave field. The wave field contains the energy loss of geometric diffusion in the process of spherical wave propagation. The cross-correlation between the source wave field and the residual backpropagation wave field leads to a more uneven energy in the deep gradient and shallow layers, which will cause insufficient accuracy of the inversion velocity in deep layers.

However, the calculation of the Hessian matrix is ​​accompanied by a huge amount of storage and calculation. How to avoid the calculation of the Hessian matrix or further approximate the Hessian matrix is ​​a problem that needs to be considered

Images