Use of graphical structures to quantify and propagate uncertainties in development of conceptual models
F.A. Kerl, A. Sharif Heger, D.P. Gallegos, Pam Davis · OSTI OAI (U.S. Department of Energy Office of Scientific and Technical Information) · 1991
Part of the application for a license for a high-level radioactive waste repository is an assessment of repository performance over thousands of years, which will inevitably have to treat uncertainties. One source of uncertainty is in the conceptualization of the natural repository system. Generally, conceptual models are developed based on interpretation of existing data using expert judgment. Uncertainties in conceptual models, which are propagated through the performance assessment calculations, are introduced when simplifying assumptions are made about the behavior of the real system. These assumptions are made because the data, knowledge that is based on the data, and other information considered in the interpretation are incomplete. Additionally, any relationships that exist among the data are generally inexact or may be undefined. In this work, causal networks have been applied to the conceptual model development process. This representation of the conceptual model expresses existing knowledge about the real system in a graphical form and extracts the qualitative dependency relationships among the underlying data and assumptions. Strict probabilistic reasoning is used to quantitatively explore these relationships. This probabilistic network provides a means by which to quantify, propagate, and reduce the pervading uncertainty in a coherent probabilistic manner. The conceptualization of the Avra Valley regional ground water flow system in Arizona and the ground water flow system of the proposed high-level radioactive waste repository site at Yucca Mountain in Nevada have been investigated to develop a preliminary data base of important assumptions and their relationships. Based on the conceptual models for these sites, a prototype version of the probabilistic network for the development of conceptual models is under development on a microExplorer Lisp workstation.