Reservoir simulation is essential in the decision-making process for the development and management of petroleum reservoirs. A simulation model can predict the reservoir behavior under various operating conditions. Hence, engineers can test different locations for infill-drilling wells or investigate the performance of an enhanced oil recovery process, for example. However, the actual properties of a reservoir are poorly known. Therefore, it is essential to incorporate all relevant (and consistent) information about the reservoir in the models. The process of incorporating dynamic data into reservoir models is known in the petroleum literature as history matching.
History matching has been the subject of intense investigation and development in the last 4 decades. However, in practice, we still observe a significant number of engineers manually history matching their models in an arduous and tedious process of trial and error. Moreover, the need for uncertainty quantification demands that engineers provide multiple history-matched models, which does not make life easier. Fortunately, this scenario is gradually changing. First, advances in computer hardware and software allow engineers to run multiple reservoir simulations in a reasonable time. Second, computer-aided history-matching tools assist the process, reducing the human time spent on repetitive activities, which results in more time for analyzing results and making decisions. Nevertheless, this tale is far from the happy end. Even though we have faster computers and sophisticated assisted-history-matching methods, we always want (need) more. We want more geological realism. We want more integration among disciplines. We want to incorporate different (and sometime exotic) types of data. We want better uncertainty quantification. We want to close the loop and make decisions in real time. This makes history matching a fascinating, challenging, and prolific research area.
Perhaps it is fair to state that the concept of history matching is evolving from the idea of finding the best model (i.e., the model that best reproduces the field observations) to the idea of a process of uncertainties mitigation. In this sense, the modern interpretation of history matching is better defined as a sampling problem rather than an optimization (minimization) problem. In this interpretation, the goal is to explore the uncertainty space searching for solutions (samples) that are consistent with the geological information and able to reproduce the observations within the confidence level of the data. Interpreting the history matching as a sampling problem does not diminish the importance of optimization methods though. In this concept, optimization becomes a mathematical tool for solving the sampling problem efficiently.
The papers summarized in this feature and the ones indicated in the additional-reading list are good examples of recent developments and field applications of history matching.
Additional reading
SPE 163652 Prior-Model Identification With Sparsity-Promoting History Matching by Mohammadreza M. Khaninezhad, Texas A&M University, et al.
SPE 152805 Fast and Efficient Assisted History Matching for Large-Scale Applications by Torsten Friedel, Schlumberger, et al.
SPE 159344 Integration of 4D-Seismic Monitoring Results as History-Match Indicators for Reservoir Simulation by Amna Ali, Total, et al.
