A perturbation-based hybrid methodology for control drum worth prediction applied to the HOLOS-Quad microreactor concept

Abstract

Fast and accurate control system reactivity worth estimates are desirable in many applications. Unfortunately, the full core Monte Carlo transport simulations often used to estimate the reactivity worth of control drums are often too costly to run in the time scale needed for many applications, such as model predictive control, reactor design optimization, uncertainty quantification and sensitivity analysis. Therefore, this paper presents a general methodology for control drum worth estimation that is based on a hybrid model consisting of physics-based and statistics-based components. It was found that in the application of the methodology to the HOLOS-Quad Reactor Concept, errors as low as 50 pcm were achieved. Model accuracy was also analyzed as a function of training set size and it was found that the most accurate model required about 70 transport model evaluations to train with further reactivity estimates being calculated on the order of milliseconds.