A fast approach to ensemble appraisal and reservoir performance predictions

It is only when the underlying uncertainties in reservoir model parameters are quantified that proper reservoir management decisions can be made. In this paper, we present an approach for making posterior inferences from the ensemble of reservoir models generated during history matching. The strength of the approach lies in the fact that it is faster than the predominantly used methods. It relies on high quality proxy models developed through a Genetic Programming Based Symbolic Regression. As a result, the expense of solving the forward problem is avoided at this appraisal stage. However, the probability distribution of parameters is initially unknown so the model space is resampled systematically according to the posterior probability density function. This results in the calculation of the Bayesian statistical measures of model plausibility and the correlations of the model parameters. The effectiveness of the approach is demonstrated here on model realisations generated using a Genetic Algorithm, but it is equally applicable to models generated through any other stochastic search methods. The results suggest that the new approach is an accurate and fast alternative to the existing methodologies for ensemble appraisal and stochastic reservoir performance forecast. MCMC resampling with the proxy model takes minutes instead of hours.