Model Uncertainty Assessment; Review of Available Approaches

Model Uncertainty Assessment; Review of Available Approaches

The term “model” refers to the way that the behavior of a system or phenomenon of interest is predicted using physical, verbal, human or mathematical representations. A model is thought of as being composed of the structural elements and relations, plus a particular form reflecting how these elements and relations interact. No model could represent the reality as the exact way it exists because models are always constructed on the simplifying assumptions to make a complex problem manageable. Therefore uncertainty is unavoidable in every modeling process. This becomes more complicated as the dimensions of the problem increases. Probabilistic risk assessment (PRA/PSA), thermo-hydraulics calculations (to resolve the safety issues in nuclear industry) and environmental modeling are good examples in this regard. In general, uncertainty is involved in selecting the best model among alternative models available. This uncertainty is called model uncertainty. Two types of model uncertainty (single model vs. multiple models) will be discussed here. Although the model uncertainty is a very important source of uncertainty, it is usually ignored in the engineering applications. Model uncertainty is in its early development stages in the scientific community and there is no consensus methodology for its characterization and quantification. This paper is aimed at reviewing the subject matter to evaluate the state of the art on the topic. Different approaches proposed by various researchers for model uncertainty quantification will be presented with a concentration on the safety issues. The methodologies will be compared in order to summarize their merits and limitations.