Full Waveform Inversion Of Unwrapped Phase Based On Dual-parameter Shaping Regularization

The exploitation under the complex substratum requires a very high resolution of imaging, thus the normal processing technology should be constructed basing on the seismic migration with a high resolution, while the resolution of velocity is crucial to that of migration. In this paper, we primarily research full waveform inversion for velocity models. Contrast to the travel-time imaging methodology, full waveform inversion remedy the information under high frequency. And comparing with velocity analysis, full waveform inversion could obtain a finer substructure and have details on the situation of oil-gas storage.Full waveform inversion relies on the least-square migration,which applies the adjoint technology correlating the seismic waveform with back propagation to update the model and match the true model. This paper mainly study the improvement of objective functions and the optimize algorithm. Firstly, I proposed two points: one point is to add dual-parameter regularization into the objective function so that the boundary would be clearer; the other point is to use the unwrapped phase to enhance the resolution of FWI. Secondly, I compared both of L-BFGS and Powell algorithms in attempt to speed up the inversion and convergence.The objective function illustrates the relation of datum and models, which reflects the fundamental meaning of FWI, that is, fully using the entire formation of waveform including its phase and amplitude. Therefore, it is vertical to construct the rational objective function to inverse the real geological model. Due to the lack of information for low frequencies, the bottom of the model is very blur. Thus, the dual-parameter regularization methodology has been proposed, in which these two parameters are beneficial for balancing the smoothing and shape of imaging. Besides, the wrapped phase leads to the local matching in frequency domain, and I applied the unwrapped phase for constructing the objective function. I inferred the new function about the phase in the frequency domain to make up the lack of information for waveform. Finally, I accomplished the FWI on the Marmousi models and metal models, and the results showed the improvement of resolutionThe optimize algorithm is an important tool for inversion. To choose different algorithms depends on the convergence and computation. This paper studied Powell and L-BFGS algorithms, analyzed their accuracy and compared their results. I achieved them on Marmousi models, and separately, obtained their inversed models and curve lines representing their convergence. By means of comparison, I got the conclusion, that constraint conditions promoted convergence of inversion, and Powell algorithm has a higher speed for inversion than that of L-BFGS, but the latter is more stable than the former.