Research On Waveform Inversion In Elastic Medium

The objective of reflection seismic exploration is to obtain high-precision parameter distribution of subsurface by extracting information from seismogram. With the development of seismic exploration, higher-precision parameter distribution is required. Conventional seismic data processing method cannot meet demand anymore because of theoretical limits. Waveform inversion is a inverse problem with complete theoretical basis. It gets the subsurface parameters through non-linear inversion which objective is waveform matching between forward and observational data. As it uses all the waveform information of seismic data, such as travel time, amplitude et al, and take the full wave equation as modeling operator as well, waveform inversion has higher resolutions in theory. What’s more, the theory based on elastic medium hypothesis can describe more accurate seismic wave propagation than acoustic medium and it provides the prerequisites for multi-parameter inversion. Elastic wave waveform inversion can help make up the theoretical insufficiency of conventional seismic exploration to get higher accuracy parameter estimation and provide more geophysics evidence for reservoir description.Firstly, the paper discussed the strategy of model parameterization which is the first step of multi-parameter inversion. The different parameter selection will affect the inversion resolution of different parameters. Radiation analysis provided a theoretical basis and corresponding physical interpretation for parametric analysis, but it is based on homogeneous medium hypothesis. This paper gives some strategies of model parameterization for reflection seismic exploration in complex medium by some numerical tests.In addition, this paper presented multi-scale inversion algorithms in frequency domain and time domain respectively which can weaken the dependence of waveform inversion method to initial models. By appropriate selection of inversion frequency, we can get multi-scale inversion of low, intermediate and high wavenumber respectively. Frequency-domain waveform inversion which deals with single frequency data directly, is the best way to do multi-scale inversion, and its high efficiency in solving multi-sources problem reduce the computational cost. But this method need huge memory and computation time when deal with larger scale. This paper proposed mix-domain multi-scale inversion method where wavefield simulation is done by algorithm in time-domain and gradient is calculated in frequency-domain. This method combines the high efficiency of time-domain forward algorithm in large scale models with the advantage of frequency-domain multi-scale inversion. It’s a better way to process large scale field data. What’s more, in order to solve the problem of step length estimation for different parameter in multi-parameter inversion, this paper applied subspace searching technology to elastic waveform inversion method.It’s actually a spherical wave when seismic wave propagating in real media, so it will have higher accuracy using 3-D waveform inversion to process observational data. Considering the extensive calculation of 3-D waveform inversion, this paper presented parallel algorithm based on domain decomposition for PC-Cluster.