Bayesian Forecasting for Complex Systems Using Computer Simulators

Abstract

Although computer models are often used for forecasting future outcomes of complex systems, the uncertainties in such forecasts are not usually treated formally. We describe a general Bayesian approach for using a computer model or simulator of a complex system to forecast system outcomes. The approach is based on constructing beliefs derived from a combination of expert judgments and experiments on the computer model. These beliefs, which are systematically updated as we make runs of the computer model, are used for either Bayesian or Bayes linear forecasting for the system. Issues of design and diagnostics are described in the context of forecasting. The methodology is applied to forecasting for an active hydrocarbon reservoir.

Probabilistic Inference for Future Climate Using an Ensemble of Climate Model Evaluations

“…In fact there appear to be foundational difficulties that prevent us from expressing such a relationship in terms of operationally-defined quantities pertinent to the model itself Rougier, 2005, 2006b). These objections notwithstanding, this paper adopts the simplest framework that takes explicit account of the model's imperfections; this is also the de facto standard in the statistical treatment of model-based inference for complex systems (see, e.g., Craig et al, 2001;Kennedy and O'Hagan, 2001;Higdon et al, 2005;Goldstein and Rougier, 2006a).…”

Predicting Snow Velocity in Large Chute Flows Under Different Environmental Conditions

“…For these reasons we strongly favour a Bayes linear analysis. The Bayes linear approach is outlined in Goldstein (1999) and described in detail in Goldstein and Wooff (2007); it has proved very powerful in large computer experiments (Craig et al, 1997(Craig et al, , 2001Goldstein and Rougier, 2006). It also underpins standard techniques such as Dynamic Linear Models (West and Harrison, 1997), also known as the Kalman filter.…”

“…A particular challenge in this field is to account for the fact that some of the model parameters are imperfectly known, and that the model itself is imperfect (Kennedy and O'Hagan, 2001;Craig et al, 2001;Goldstein andRougier, 2004, 2009). This challenge becomes more acute when the model-outputs and the system behaviour are multivariate.…”

Calibration of Stochastic Computer Simulators Using Likelihood Emulation

“…Our approach is inspired by the framework for Bayesian calibration proposed by Kennedy and O'Hagan (2001) and developed in Higdon et al (2004), Bayarri et al (2007b), Bayarri et al (2007a), and Higdon et al (2008), and by Bayes' linear history matching developed by Craig et al (2001), Goldstein and Rougier (2006), and Vernon, Goldstein, and Bower (2010). We refer to a computer model as a "simulator" and focus on three issues: computationally expensive simulators; "discrepancy," which is the error in a simulator prediction due to the simulator being an imperfect model of reality; and stochastic simulators, which are simulators that can return different output values from the same input values.…”