2017/10/09 by Christopher Drovandi, Drovandi, Christopher C, David J. Nott +3 · 1 citation
Engineering · Decision Sciences · #Reservoir Engineering and Simulation Methods #Nuclear reactor physics and engineering #Forecasting Techniques and Applications
paper · pdf · doi:10.48550/arxiv.1710.03133
The aim of the history matching method is to locate non-implausible regions\nof the parameter space of complex deterministic or stochastic models by\nmatching model outputs with data. It does this via a series of waves where at\neach wave an emulator is fitted to a small number of training samples. An\nimplausibility measure is defined which takes into account the closeness of\nsimulated and observed outputs as well as emulator uncertainty. As the waves\nprogress, the emulator becomes more accurate so that training samples are more\nconcentrated on promising regions of the space and poorer parts of the space\nare rejected with more confidence. Whilst history matching has proved to be\nuseful, existing implementations are not fully automated and some ad-hoc\nchoices are made during the process, which involves user intervention and is\ntime consuming. This occurs especially when the non-implausible region becomes\nsmall and it is difficult to sample this space uniformly to generate new\ntraining points. In this article we develop a sequential Monte Carlo (SMC)\nalgorithm for implementing history matching that is semi-automated. Our novel\nSMC approach reveals that the history matching method yields a non-implausible\nregion that can be multi-modal, highly irregular and very difficult to sample\nuniformly. Our SMC approach offers a much more reliable sampling of the\nnon-implausible space, which requires additional computation compared to other\napproaches used in the literature.\n