The initial work on computer techniques for assisted history matching date back to the 1960s. However, it was a long journey between the development of early methods and operational use. Initially, these methods were referred to as “automatic history matching,” giving the wrong impression that it was something we could delegate to a computer. Fortunately, industry and academia soon realized that “assisted” was a more accurate term than “automatic.”
Nowadays, there is an impressive amount of literature and a large number of assisted-history-matching methods. The diversity is so vast that it is challenging to divide these methods into categories. For example, there are several flavors of methods based on sensitivity and gradient-optimization algorithms. There are also methods grounded on stochastic optimization, evolutionary algorithms, design of experiments, proxy modeling, streamline simulation, and Kalman filters. This is just to mention a few, and the list is still growing.
Regardless of the specific method of your choice, I believe the main development in history matching is the recognition that it can be formulated as a Bayesian inference problem. Bayes’ rule provides an elegant framework to formalize the process of learning from data to update our beliefs. The beauty is in the fact that Bayes’ rule gave to history matching the correct meaning. History matching is no longer a searching process to find the best model. Instead, history matching is a process of mitigating uncertainty in light of new information.
The papers summarized in this feature and the ones indicated in the additional-reading list are excellent examples of recent developments and applications of assisted-history-matching techniques. All are aligned with the modern Bayesian interpretation. I hope you enjoy the reading.
This Month's Technical Papers
Uncertainty Quantification for History-Matching Problems
Field-Scale Assisted History Matching Using a Systematic Ensemble Kalman Smoother
Drill and Learn: A Decision-Making Work Flow To Quantify Value of Learning
Recommended additional reading
SPE 179549 Streamline-Based Rapid History Matching of Bottomhole Pressure and Three-Phase Production Data by Dongjae Kam, Texas A&M University, et al.
SPE 182684 Generation of a Proposal Distribution for Efficient MCMC Characterization of Uncertainty in Reservoir Description and Forecasting by Xin Li, The University of Tulsa, et al.
SPE 182693 A Robust Iterative Ensemble-Smoother Method for Efficient History Matching and Uncertainty Quantification by Xiang Ma, ExxonMobil, et al.
