In engineering one often have to deal with uncertainty while making decisions. There can be uncertainty about the current status of a system, as well as future trends. In this talk we will explore how Equinor deals with uncertainty in reservoir planning. We will see how one can optimize over an uncertain model and how production data is utilized to improve our models for future decisions. In the end we will see how the same knowledge and tools can be utilized to better understand windmill farms and how they should be designed.

The talk will give an introduction to ensemble based methods, as well as the process of optimizing over and updating such models. We will explore variants of the Kalman Filter algorithm that can be utilized for ensembles of models and how gradients can be efficiently approximated. It will all be given in a practical setting with concrete examples. No prior knowledge is required (Bayesian pun intended).

Both the optimization and the model updating methods are implemented in software projects and in active use in Equinor. The model updating tool is an open source project named ERT (github.com/equinor/ERT) and has been used outside of Equinor by among others Norsk Regnesentral, TNO and Norce.