A localised sensitivity analysis (SA) is presented to enhance the 4D seismic history matching (4D SHM) parameterization. It helps find all the influential parameters and avoid wrongly excluding some important parameters from the 4D SHM process. We apply different localisation approaches such as a sliding window to scan the entire 4D seismic map and dimensionality reduction and feature extraction methods to extract the information content of the 4D seismic maps. The latter transforms the 4D seismic maps into a latent space and represents the maps in lower dimensions. We argue that the dimensions in the latent space are representative of the main signals on the 4D seismic maps. Hence, doing an SA to these features is equivalent to doing it for the main signal in the original space. We couple this scheme with ensemble smoother with multiple data assimilation (ESMDA) to perform different 4D SHM scenarios based on the conventional and localised SA outcomes. We see that localised SA is superior to conventional SA in terms of its impact on 4D SHM results. By comparing the sliding window, principal component analysis (PCA), autoencoders, and variational autoencoders, it turned out that PCA is the most suitable tool for localising the SA.
|Number of pages||5|
|Publication status||Published - 5 Jun 2023|
|Event||84th EAGE Annual Conference & Exhibition 2023 - Vienna, Austria|
Duration: 5 Jun 2023 → 8 Jun 2023
|Conference||84th EAGE Annual Conference & Exhibition 2023|
|Period||5/06/23 → 8/06/23|