| Title | Ensemble methods for geothermal model calibration |
|---|---|
| Authors | A. de Beer, M. Gravatt, R. Nicholson, J.P. OSullivan, M.J. OSullivan, O.J. Maclaren |
| Year | 2023 |
| Conference | New Zealand Geothermal Workshop |
| Keywords | reservoir modelling, model calibration, uncertainty quantification, ensemble methods |
| Abstract | A typical geothermal model requires significant computational resources to simulate and can contain hundreds of unknown parameters. The process of estimating these parameters, often referred to as model calibration, is a difficult task; traditional methods such as Markov chain Monte Carlo generally require running a prohibitively large number of simulations to obtain accurate results. Ensemble methods form an alternative class of algorithms for approximating the solution to the calibration problem and have the potential to provide accurate results using considerably fewer simulations. Ensemble methods have been used successfully to calibrate large, complex models in areas including petroleum engineering, oceanography, and weather forecasting. There are, however, few examples of applications of these methods to geothermal reservoir modelling. In addition, the wide variety of ensemble methods that have been developed mean there is a need for numerical studies that examine their respective benefits and drawbacks when applied to specific problems. To support the effective use of ensemble methods for geothermal reservoir model calibration, we review two widely used ensemble methods and apply them to the problem of calibrating a synthetic reservoir model. We demonstrate that both methods are capable of generating accurate reconstructions of the model parameters, with appropriate characterisation of uncertainty. |