CDI | Accurate Models Delivered Ten Times Faster than the Original Model Building Project
The first test of hyperModel used a 500 sq. km data set from Côte d’Ivoire, West Africa, and benchmarked hyperModel with a full-integrity tomographic model build on the same area.
The hyperModel workflow was used to build a velocity model globally, reducing turnaround by removing intervention. Two initial models were tested, first using the starting model for the full-integrity tomographic model building, and secondly with this initial model modified to incorporate locally varying errors of up to 15%. The two outputs from hyperModel were correlated with the final tomographic model generated using the same data.
The slider shows the initial hyperModel inputs (left) and the outputs (right) of three sets of Common Image Gathers (CIGs) and three models co-rendered on seismic data. E is the initial model used for the full-integrity tomography and B the gathers that result from a migration with that model. F is the modified initial model and C the migrated CIGs. The gathers in C are both under and overcorrected, and indicate that the model is very inaccurate. The model in D is the result from the full-integrity tomographic model build. The gathers are flat (A), and the model locally conformable to the geology.
Input and output model for the Campos Basin phase 1 hyperModel work. Note the geological consistency of the model, capturing regional changes in velocity and flattening both post and presalt sequences.
The two outputs from hyperModel are comparable to the full-integrity model (A and D are the same as the previous figure). The CIGs in B and C, display a similar level of moveout as A and the gathers are generally flat. Despite starting from a much less accurate model, the gathers in panel C are very similar to B, and the two hyperModel outputs (E and F) are almost identical. Both models look comparable to the full-integrity model in D, with one notable exception annotated by the orange and blue arrows. In the full-integrity velocity model building, high velocity channels benefited from geobody interpretation and masked updates (orange arrow). During the model building process this was the only way to focus the update, however, the same channel is partially updated using the Monte Carlo simulation (blue arrows). nnel is partially updated using the Monte Carlo simulation (blue arrows).
Building a velocity model using an automated Monte Carlo simulation converges on solutions of equal quality to a full-integrity model build. In this field example, the starting point was up to 15% in error but still resolved. The original model building project took 90 days, however, both hyperModels delivered accurate models in less than an order of magnitude of that time.