Integrating prior knowledge through multiple kernel learning for the prediction of petroleum reservoir properties

Prediction of petroleum reservoir properties such as permeability and porosity can be enhanced by including prior knowledge in the form of multiple possible geological scenarios. The difficulty in integrating prior beliefs is in the diversity of the information available. This knowledge is often conflicting, heterogeneous and scale dependant. In this paper we show how Multiple Kernel Learning (MKL) uses data fusion to integrate prior knowledge in the form of possible prior geological scenarios to predict the spatial distribution of permeability & porosity. MKL
integrates the data in an intelligent way, preserving the complex relationships between the prior knowledge and the predictions with the use of individual kernels. The proposed approach addresses the problem of uncertainty of the geological scenarios created based on sparseness of data and various modelling assumptions.