Loading...
Please wait, while we are loading the content...
Similar Documents
Multiscale Full Waveform Inversion Using a Filtering Approach and Gradient Preconditioning
| Content Provider | Semantic Scholar |
|---|---|
| Author | Cristo, Rafael De Porsani, Milton |
| Copyright Year | 2018 |
| Abstract | ABSTRACT. The FWI multiscale approach in data domain produces better results because the problem gets closer to the overall minimum avoiding the local minima. The method works in different scales, avoiding the initial velocity model choice as well as the cycle skipping. Regarding to multiscale approach, it was done choosing frequencies band performed by Wiener filter and a SVD filter trace by trace both in data domain. The trace by trace SVD filter works taking each trace of the gradient and assembles on the shifted matrix traces and do the decomposition from low to high frequency. In addition this multiscale approach in data domain was compared to another multiscale approach using damping filters on the objective function (MDFOF). Due to the problem of geometrical spreading, during the propagation of the wave field, the deeper regions of the model are not well illuminated, hence the preconditioning of the objective function gradient was done in order to eliminate this problem and allow the deeper regions to be compared. Keywords: SVD filter; Full waveform inversion; Gradient preconditioning; Pseudo Hessian diagonal. RESUMO. A abordagem multiescala no problema da FWI, produz melhores resultados pois o problema consegue convergir para o minimo global, evitando o problema do minimo local. O metodo funciona em diferentes escalas, evitando o problema da escolha no modelo inicial de velocidade bem como o problema de salto de ciclo. Em relacao a abordagem multiescala, o mesmo foi realizado escolhendo bandas de frequencias usando o filtro de Wiener e o filtro SVD traco a traco. O filtro SVD traco a traco funciona tomando cada traco do gradiente e da matriz de tracos deslocados a faz a decomposicao das baixas as altas frequencias. Alem dessa, abordagem multiescala no dominio do dado, outra abordagem multiescala usando filtros atenuantes foi comparada com a abordagem multiescala no dominio do dado. Devido ao problema de divergencia esferica, durante a propagacao da onda, as regioes mais profundas do modelo nao sao corretamente imageadas, portanto faz-se necessario o precondicionamento do gradiente foi feito com intuito de eliminar esse problema e permitir a comparacao das duas abordagens nas regioes mais profundas do modelo. Palavras-chave: Filtro SVD; Inversion complete da forma de onda; Precondicionamento do gradiente; Diagonal da Pseudo Hessiana. |
| Starting Page | 247 |
| Ending Page | 262 |
| Page Count | 16 |
| File Format | PDF HTM / HTML |
| DOI | 10.22564/rbgf.v35i4.899 |
| Volume Number | 35 |
| Alternate Webpage(s) | https://sbgf.org.br/revista/index.php/rbgf/article/download/899/932 |
| Alternate Webpage(s) | https://doi.org/10.22564/rbgf.v35i4.899 |
| Language | English |
| Access Restriction | Open |
| Content Type | Text |
| Resource Type | Article |
