Глобальная ядерная безопасность (Apr 2023)

Application of hierarchical agglomerative clustering to create a basic classes library of a VVER 1000/1200 reactor facility

  • Gennadiy V. Arkadov,
  • Irina V. Trykova,
  • Denis V. Zvyagincev,
  • Konstantin I. Kotsoev

DOI
https://doi.org/10.26583/gns-2023-01-05
Journal volume & issue
Vol. 0, no. 1
pp. 54 – 66

Abstract

Read online

In reactor plants with a water-water power reactor (VVER), free, weakly fixed and foreign objects may appear in the main circulation circuit, posing a threat to the integrity of the equipment and the safety of the reactor plant. For the purpose of early detection of these objects, the NPP is equipped with a system for detecting loose/weakly fixed objects (SOSP).In addition to the detection of loose/weakly fixed objects, the functions of the SOSP include the classification of registered events.The possibility of applying the classification algorithm is based on the fact that the signals from the operation of standard equipment are highly repeatable, even in the presence of noise, while a free object is characterized by a large stochastic component and its own deterministic class cannot be formed for it.Classification reduces the number of false alarms, allowing you to select signals from regular operations, while signals from one process must be assigned to one class.The idea of ​​the article is to "train" SOSP on a certain archive of data characterizing the normal functioning of the reactor plant, create a library of "base" classes and set the boundaries of each class so that, on the one hand, take into account the possible variability of signal parameters due to noise.Having defined the base classes, we can state that if a newly received signal falls into one of the classes, then it reflects the regular operation of the RI, while signals that do not fall into any of the classes may be the result of the appearance of a free/weakly fixed object.The article analyzes a lot of events accumulated in the archive of one of the existing SOSP.Their clustering was carried out, as a result of which the classes of events corresponding to regular technological operations were identified.For each class, the center of the class and the allowable limits of deviations from the center are calculated.All class centers obtained are benchmarks against which the SOSP either classifies a newly detected event in real time or characterizes it as "unclassified".

Keywords