Title: Hierarchical compression-based reservoir modelling conditioned to seismic and well data
Researchers: Deirdre Walsh, Dr Tom Manzocchi
The ability to generate geologically realistic reservoir models that honour the available well data is an important step in predicting reservoir behaviour. Many deep marine turbidite reservoirs are often characterised by poorly amalgamated sand bodies interbedded with low permeability shales. Although these systems often have high net:gross ratios (NTG), the low connectivity of the sandstones has a strong control on reservoir performance but is often poorly reproduced in reservoir geomodels. A new class of object-based model, which uses the so-called compression method in order to reproduce poorly amalgamated but high NTG sequences, was developed by Manzocchi et al. 2007. The compression method allows NTG and amalgamation ratios (AR) to be separate inputs into the modelling workflow. Both standard and compression-based object modelling methods struggle to honour available well and seismic data. The objective of this project is to address this problem by combining the compression algorithm (Manzocchi et al. 2007) with the multi-point statistics (MPS) method (Strebelle 2002). This will require a new workflow that generates MPS models using a “decompressed” object-based model as a training image and “decompressed” wells as conditioning data. The “decompressed” training image represents target connectivity with a reduced NTG. The compression algorithm is then applied to the MPS model to provide the final model which will include appropriate levels of amalgamation yet will honour the well data.
The objectives of the PhD project are as follows:
1. The basic workflow needed will be defined and tested, with the existing in-house code modified to provide the necessary pre-and post- processors to Petrel in which the MPS modelling will be performed.
2. The success of the output models will be tested by comparing static connectivity and dynamic flow characteristics of the output MPS models against the input training models.
3. The method will be tested using more sophisticated conceptual reservoir models, for example including hierarchically constructed training image.
4. The method will be refined to include “soft” seismic as well as “hard” well conditioning data.
5. Test the modelling in a real sub-surface reservoir case-study.