Many reservoir engineers dislike the very idea of automatic history match- ing applied to real full-field studies. They believe there is no artificial substitute for experienced reasoning, deep understanding of the reservoir mechanisms, and atten- tion to real-life practical aspects of the problem. Some use terms such as art and intu- ition. For them, even if computers long ago learned to play chess, computers will never be able to perform real-case history matching on their own or at least they are still too far from this achievement. Very often, during technical sessions, immediately fol- lowing an advanced mathematical presentation on history matching, someone in the audience makes his or her point about the limits of automatic approaches. To avoid disputes, experienced speakers prefer less pretentious expressions such as assisted or semiautomatic history matching.
Indeed, history matching can be seen as a two-step iterative process, normal- ly requiring many cycles to be completed. Broadly speaking, the first step is about analysis and setting the problem parameters, and the second step is about search- ing for and computing solutions. We start our discussion with the second part, which has a more obvious algorithmic nature. There has being a great deal of research and progress in this area. The ensemble Kalman filter is dominating the scene, but gradi- ent-based methods and global-optimization stochastic methods are attracting mer- ited attention. Most published contributions come from universities, and, typical- ly, papers include examples to demonstrate successful algorithm application. These examples can be simple synthetic or somewhat-more-realistic cases, but the discus- sion is naturally focused on the solution method and not on the entire problem as found in the field.
The first part of the problem is less mathematized, for now, and involves essen- tial tasks such as to be clear about the practical purposes and requirements in the par- ticular context; to have a full understanding of the quality of the reservoir model and the production data; to design or redesign well-justified objective functions; to set adequate parameterization, considering the main uncertainties and their effect on the simulation results; to represent properly and sample the uncertainty space; and to evaluate results from the previous steps of the history-matching process judiciously. Unfortunately, the strategies used to consider this part of the problem are much less discussed and documented. In fact, many of these tasks are open to further formal- ization and, ultimately, can be automated also. We definitely need more papers illu- minating these other aspects of the reservoir-engineering problem, instead of relying on intuition.
Read the paper synopses in the April 2012 issue of JPT.
Régis Kruel Romeu, SPE, is a Senior Consultant at Petrobras Research Center (CENPES) in Rio de Janeiro. With 31 years’ experience in petroleum engineering, he has worked mostly in reservoir- characterization and -simulation applied research. Romeu’s main activities and areas of interest are heterogeneities representation, scale up, history matching and optimization, integrated reservoir studies, coordination of research projects, relationship with Brazilian universities, and reservoir studies related to Brazilian presalt fields. He holds a BS degree in civil engineering from Universidade Federal do Rio Grande do Sul, Brazil; an MS degree in petroleum engineering from Universidade Federal de Ouro Preto, Brazil; and a PhD degree in quantitative geosciences from the Université Pierre et Marie Curie, Paris. Romeu has served as Editor for SPE Res Eval & Eng and serves on the JPT Editorial Committee.