Deep learning techniques for in-core perturbation identification and localization of time-series nuclear plant measurements

The use of machine learning in the field reactor safety and noise diagnostics has recently seen great potential given advancements made computational tools, hardware simulations. In this work we demonstrate how deep neural networks, specifically recurrent convolutional networks can be trained a synthetic setting aligned to operate on real plant measurements recover perturbation type origin location from time-series signals. We first utilize vast quantities data generated extended SIMULATE-3K codes, simulating Swiss 3-loop pre-KONVOI train our under variety differing settings. Additionally, extend these approaches unsupervised measurements, where information about true characteristics is unknown. As such, show applicability self-supervised domain adaptation approach correctly align representations learned by network between both detector readings more concretely classify localize perturbation. validate number experimental analyses showing successful performance simulated domains.