Learning model discrepancy: A Gaussian process and sampling-based approach

Predicting events in the real world with a computer model (simulator) is challenging. Every simulator, to varying extents, has discrepancy, mismatch between observations and simulator (given ‘true’ parameters are known). Model discrepancy occurs for various reasons, including simplified or missing physics numerical approximations that required compute outputs, fact assumptions not generally applicable all contexts. The existence of problematic engineer as performing calibration will lead biased parameter estimates, resulting unlikely accurately predict (or even be valid for) contexts interest. This paper proposes an approach inferring overcomes non-identifiability problems associated jointly along discrepancy. Instead, proposed procedure seeks identify given some distribution, which could come from ‘likelihood-free’ considers presence during calibration, such Bayesian history matching. In this case, inferred whilst marginalising out uncertain outputs via sampling-based approach, therefore better reflecting uncertainty Verification performed before demonstration on experiential case study, comprising representative five storey building structure.