9311950 Shortliffe This research addresses the problem of modeling time in dynamic domains given incomplete and uncertain information about the domain. The dynamic network model (DNM) is a time-series model built on the belief-network paradigm that employs classical time- series methodologies to provide a mechanism for making unbiased forecasts. Domains exhibiting complex cyclical behavior abound in all fields of science and engineering. The success of dynamic models in these application domains is contingent on their ability to predict the future cyclical behavior. Typically, these domains contain uncertainty in the form of noise from perturbations by unmodeled exogenous forces. An integrated framework will be provided for modeling cyclical behavior an uncertainty. The methodology combines Fourier analysis, a powerful technique for modeling periodic functions, and belief networks, an expressive knowledge representation for modeling uncertainty.
