A continuous learning monitoring strategy for multi-condition of nuclear power plant. (15th December 2021)

• Record Type: Journal Article
• Authors: Yu, Yue; Peng, Min-jun; Wang, Hang; Ma, Zhan-guo; Cheng, Shou-yu; Xu, Ren-yi
• Is Part Of: Annals of nuclear energy. Volume 164(2021)
• Publication Date: 2021-12-15
• Subjects: Random forest -- Continuous learning -- Principal component analysis -- Multiple operation condition; Nuclear energy -- Periodicals; Nuclear engineering -- Periodicals; 621.4805
• DOI: 10.1016/j.anucene.2021.108544
• Languages: English
• ISSNs: 0306-4549

Abstract

Highlights: Sensor fault and system fault discrimination technology based on reconstruction. Multi-condition discrimination technology of nuclear power plant based on random forest. Random forest continuous learning strategy based on principal component analysis "supervision". Abstract: Safety is the most important feature in the operation of nuclear power plants. developing and perfecting energy markets lead to an unavoidable problem of adjusting the structure of the random forest model to adapt to new situations, such as switching to other steady state operating conditions of nuclear power plants. Another critical challenge is how to effectively learn new knowledge from continuously collected measurements and how to integrate new information into current random forest models. So, this paper proposes a continuous learning monitoring strategy for multi-Condition of nuclear power plant. Next, real data from nuclear power plants are used for modeling and the simulator is used to insert the fault test. Finally, simulation tests show that the model can effectively detect the typical faults of nuclear power plants under multiple operation conditions, and the model can effectively learn new knowledge.