Use of Machine Learning in Petroleum Production Optimization under Geological Uncertainty

Geological uncertainty is of significant concern in petroleum reservoir modeling with the goal of maximizing oil production. Stochastic simulation allows generating multiple reservoir models that can be used to characterize this uncertainty. However, the large computation time needed for flow simulation (e.g., for use in production forecasting) impedes the evaluation of flow on all reservoir models. In addition, performing a formal optimization of the well controls to maximize say NPV leads to hundreds or thousands of function evaluations, each of which requires tens to hundreds of reservoir simulations depending on the number of reservoir models available. In this work we apply machine learning techniques to provide computational savings on two fronts. We use kernel k-means clustering to select a small representative set of earth models that characterize the geological uncertainty so as to reduce the number of simulations for each optimization function evaluation, and use a kriging surrogate in the optimization to reduce the required number of function evaluations.