Applications of Sample Based Inference in Geothermal Reservoir Modelling

Abstract

The aim of this work is to develop a method for automating the calibration of numerical models of geothermal fields, within the framework of Bayesian inference. Unlike the traditional optimization techniques that obtain a single point estimate of the unknown parameters, our approach summarizes all the feasible parameters that consistent with the field data through the posterior distribution. Markov chain Monte Carlo (MCMC) sampling, Metropolis-Hastings algorithm in particular, is used to draw samples from the posterior distribution. Then, answers to the calibration problem such as parameter estimation, derivation in the model predictions, and model reliability can be given by estimating the expected values of statistics of interest over these samples.

We apply this sample based approach to calibrate a simple single-layer model of the feedzone of a well, using discharge test measurements of flowing enthalpy and pressure. This approach is able to make accurate model predictions and quantify the uncertainty in the prediction. Based on the same data set, sample based approach is compared to the optimization package ITOUGH2, and shows advantages for robust parameter estimation and accurate uncertainty quantification.