ENOS Public report 2.4 - CO₂GeoNet

In this report, we introduce a new approach to Bayesian experimental design. The proposed technique utilizes PCE for averaging the KL-divergence with respect to the prior distribution of model parameters and measurement errors. The result of this procedure is a PCE response surface for the expected information gain. The proposed technique provides dramatic acceleration of solution for optimal parameters of data acquisition processes.

The proposed PCE approach has a high degree of flexibility and can be naturally extended to other systems including simulations of CO2 sequestration process. The only assumptions that have been made include normal prior distribution and normal distribution of measurement errors. Both of these assumptions are natural and valid for a wide variety of practical systems. In addition to that, Rosenblatt transformation [Rosenblatt, 1952] can be applied to build a normal distribution in the parameter space of concern. Therefore, the proposed PCE technique can be applied to vast range of problems including CCS.

Download the pdf