A new methodology for diagnosis system with ‘Don’t Know’ response for Nuclear Power Plant

• Published in: Annals of nuclear energy 2017-02, Vol.100, p.91-97
• Main Authors: dos Santos Nicolau, Andressa, Schirru, Roberto
• Format: Article
• Language: English
• Subjects: Accidents; Classification; Evolutionary algorithms; Methodology; Nuclear engineering; Nuclear power generation; Nuclear power plants; Nuclear reactors
• ISSN: 0306-4549; 1873-2100
• DOI: 10.1016/j.anucene.2016.10.018

•New methodology for diagnosis system with “Don’t Know” response.•Here a new method is proposed to find the classification of an anomalous event within signatures of a set of design-basis accidents and normal state of a Brazilian pressurized power reactor.•Quantum evolutionary algorithm was used, being responsible for finding the representative vector of each accident class.•The simulation results show that the method proposed is able to identify the reference accident and distinguish the unknown types with different datasets. In Nuclear Power Plants, recognizing the type of accident during early stages, for taking appropriate actions, is critical. Moreover, classification of a novel accident as “Don’t Know”, if it is not contained within its accumulated knowledge, is necessary. To fulfill these requirements this article presents a new methodology for diagnosis system with “Don’t Know” response. The method proposed aims to classify an anomalous event within signatures of a set of design-basis accidents and normal state of a Brazilian pressurized power reactor, besides generating a ‘Don’t Know’ answer to accidents outside the training scope. For this purpose, quantum evolutionary algorithm was used as a method of separation of classes, being responsible for finding the representative vector of each accident class. The “Don’t Know” methodology proposed is based on nearest neighbor theory of Voronoi Diagrams, which is responsible to determine the “influence areas” around the representative vectors found by quantum evolutionary algorithm. The simulation results show that the system is able to identify the reference accident and distinguish the unknown types with different datasets. Moreover, it shows a promising way in determining “influence area” for pattern classification problems, specifically for the accident identification problem in the nuclear engineering area.