Due to the sparsity of knowledge available, there exist a variety of possible geological scenarios to describe a reservoir. A single scenario encapsulates a specific geological concept but is limited to represent the full range of geological variability. In order to reflect the geological uncertainty it is necessary to integrate a variety of these scenarios. Therefore a model based on the joint integration of multiple geological scenarios is more robust and provides more reliable predictions of reservoir properties.
Two challenges arise from integrating multiple geological scenarios into a prediction model. Firstly, a subset of scenarios has to be determined that explain the geological variability. The size of the subset should account for the geological uncertainty whilst maintaining computational efficiency and interpretability. That is, it is easier to understand and conceptualize the impact of 10 variables in a model rather 100 variables.
Secondly, these geological scenarios are most likely heterogeneous, scale dependant and represent a specific spatial region. Integration of this data is mathematically complex. It is challenging as the relationship between each input scenario and the predictor is most likely non-linear and unique.
In this paper we present methods of selecting a relevant subset of geological models then demonstrate how Multiple Kernel Learning (MKL) is used to integrate multiple scenarios. MKL balances extracted spatial features in a non linear kernel model. MKL has the advantage over existing methodologies as it preserves the unique relationship between each input scenario and the predictor.
We demonstrate this approach by modelling the permeability and porosity in the Brugge reservoir case study. Given 104 possible geological scenarios, we select a subset and integrate these with dynamic production data. We apply unique kernels to each scenario and show, through the use of history matching techniques, how this improves both the prediction of the reservoir properties as well as the quantification of the uncertainty.