Automated reasoning with dynamic event trees: a real-time, knowledge-based decision aide

Abstract

Abstract The models and data contained in a PRA Event Sequence Analysis represent a wealth of information that can be used for dynamic calculation of event sequence likehood. In this paper we report a new and unique computerization methodology which utilizes these data. This sub-system (reffered to as PREDICTOR) has been developed and tested as part of a larger system. PREDICTOR performs a real-time (re)calculation of the estimated likelihood of core-meltsas a function of plant status. This methodology uses object-oriented programming techniques from the Artificial Intelligence discipline that enable one to codify event tree and fault tree logic models and associated probabilities developed in a PRA study. Existence of off-normal conditions is reported to PREDICTOR, which then updates the relevant failure probabilities throughout the event tree and fault tree models by dynamically replacing the ‘off-the-seel’ (or prior) probabilities with probabilities based on the current situation. The new event probabilities are immediately propagated through the models (using ‘demons’) and an updated core-melt probability is calculated. Along the way, the dominant non-sucess path of each event tree is determined and highlighted.