A novel stochastic framework is described that facilities automatic history matching and uncertainty quantification workflows. The underlying algorithm of the framework combines Design of Experiments (DoE) and Markov Chain Monte Carlo (MCMC) inference techniques. A generic singular-value decomposition technique creates a Response Surface Model (RSM) from a DoE-based selection of simulations. Ensuing RSM serves as a rapid proxy to full-physics reservoir simulation. The data integration and uncertainty quantification framework extends current working practice by sampling Bayes' posterior parameter probability distribution with a rigorous MCMC method using the Metropolis-Hastings algorithm. This improves uncertainty estimates in the history-matching mode, and provides a unified platform for scenario management and analysis in the forecast uncertainty quantification mode.

The stochastic framework is tested for robustness and efficiency on a real field case. A history matching study is carried out for a complex deepwater turbidite reservoir involving multiple geologic realizations. Within the context of the study, the stochastic framework delivered (a) a complete description of the high-impact model parameters dominating the quality of the match, (b) multiple models honoring the historical production data, (c) reduced uncertainty ranges for the history-matching parameters, and (d) a quantitative measure of reliability for different measurement types. The outcome of the multi-realization history matching study reveals a better reservoir connectivity and an increased net-sand volume compared to the maximum a-posteriori (MAP) prediction attained by a commercial assisted history-matching software. A well-by-well analysis of the most probable model reveals that the match is comparable or superior to the ones delivered by the commercial tool. The field test outcome demonstrates the reliability, expediency, rigor, and computational efficiency of our stochastic history-matching and recovery forecasting framework.

You can access this article if you purchase or spend a download.