![]() |
Dakota Reference Manual
Version 6.15
Explore and Predict with Confidence
|
Epistemic uncertainty is uncertainty due to lack of knowledge.
In Dakota, epistemic uncertainty can be characterized by interval- or set-valued variables (see relevant keywords below) that are propagated to calculate bounding intervals on simulation output using interval analysis methods. These epistemic variable types can optionally include basic probability assignments for use in Dempster-Shafer theory of evidence methods. Epistemic uncertainty can alternately be modeled with probability density functions, although results from UQ studies are then typically interpreted as possibilities or bounds, as opposed to a probability distribution of responses.
Through nested models, Dakota can perform combined aleatory / epistemic analyses such as second-order probability or probability of frequency. For example, a variable can be assumed to have a lognormal distribution with specified variance, with its mean expressed as an epistemic uncertainty lying in an expert-specified interval.