Quantifying Drilling Risk Using a Monte Carlo Framework

Abstract

Our present ability to explore, develop, and monitor geothermal energy resources is hindered by the complexity of the geologic systems that host these resources. In order to improve decision making about where to drill geothermal wells, we apply a novel approach called Bayesian Evidential Learning to quantify uncertainty of temperature predictions for a prospective well using data from a nearby well. In this method, the relationship between data and prediction variables is learned by generating a training set of data using Monte Carlo simulation, which is then used to sample posterior temperature predictions without any explicit model inversion. We present results for different locations of the prospective temperature well, illustrating the spatial extent that the observed data reduces uncertainty on the prediction. This methodology is both practical and broadly applicable to other problems requiring full uncertainty quantification using spatially complex numerical models.