f The Limits of Automatic History Matching

After thirty-plus years of development of commercial reservoir simulators, and twenty-plus years of research into history matching, manual and otherwise, we continue to be surprised by what our new wells encounter in the dynamically changing subsurface, and by what our old wells produce. What are the limits of what is achievable by automatic history matching? One approach to this question is to posit the existence of an infallible automatic history matcher of some description, and then to consider what the implications would be for oilfield operational practice. The author looks at which ways in which ostensibly revolutionary technological breakthroughs are actually adopted and normalised by practicing engineers, and the long-term implications for the delivery of their early promise of savings in time, money or skilled labor. Several examples are introduced from the recent history of other information-driven industries, and from the author's field experience in the delivery and application of predictive reservoir models in the different phases of the reservoir lifecycle. In many cases, so-called "automatic" history matchers find their greatest utility not as black boxes that deliver a perfect model, but as guides to the intelligent use of more conventional manual matching techniques. What are the differences between a "perfect matcher" and a "helpful matching assistant"? Can both these design goals be achieved in a single piece of software, or is a different architectural approach required? Finally, the author speculates about the implications of these findings for the future of reservoir modelling practice, and considers how the non-specialist might be better served by the technology providers.<br>