Mathematical models are subject to variabilities, uncertainties and errors, which reduce the confidence that engineers have in using them. Model uncertainty analysis is largely restricted to the statistical analysis of parameter variability propagation and relevant methods are surveyed. However, uncertainty due to model structure and form is rarely acknowledged. A methodology utilising Bayesian Belief Networks is proposed for assessing confidence in model output. A topology is presented that captures all uncertainties associated with modelling system behaviour. Bayesian inference enables significant uncertainties within a network of models to be identified, thus allowing resources for model improvement to be better targeted.
展开▼
