Estratégias de inversão de multiparâmetros utilizando a equação completa da elastodinâmica

This thesis proposes a new methodology for the update directions for the elastic multiparameters full waveform inversion (FWI), by using the approximate Hessian reparametrization properties, to ensure that the parameters update directions are properly scaled, and also keeping proper physical units of the problem. This study also includes applications of the following local optimization methods to the 2D isotropic acoustic case: steepest-descent, two conjugated-gradients (PolakRibi`ere and Fletcher-Reeves) and the l-BFGS (Quasi-Newton). This analysis reinforces that the l-BFGS method, which makes Hessian approximations iteratively, is the one that better estimates the resulting model, besides converges faster to a satisfactory model. Also, it is evaluated the impact of the velocity models resulted from the acoustic inversions by using acoustic or elastic synthetic data. The results shows that the acoustic inversion is not sufficiently accurate to adjust the amplitudes of elastically generated data. Finally, it is explored the use of 3C/2D multicomponent (hydrophone and vertical and horizontal components of geophone) with elastic FWI. It is analyzed the effects on the inversion model results depending on the input data combination. It is concluded that whereas the input data to the FWI is elastic, the hydrophone only is the best choice if the inversion is monoparameter to the acoustic compressional velocity, and the use of all components registered are recommended when the inversion is elastic multiparameter. Furthermore, the obtained models are more accurate when the multiparameter inversion is considered.

Grantor dc:publisher

Universidade Federal do Rio de Janeiro

Year dc:date.issued

2017

Author dc:creator

• Carneiro, Marília da Silva Ramos

Advisor dc:contributor.advisor

• Landau, Luiz

• Engenharia civil
• Inversão sísmica
• Elasticidade

Statement dc:rights

• Acesso Aberto

Language dc:language

por

Handle dc:identifier.uri

http://hdl.handle.net/11422/7221

OAI identifier oai:identifier

oai:pantheon.ufrj.br:11422/7221