Transient trend prediction of safety parameters for small modular reactor considering equipment degradation. (February 2023)

• Record Type: Journal Article
• Authors: Zhang, Bowen; Wang, Shibo; Cheng, Shouyu; Sun, Jilin; Peng, Minjun; Wang, Chenyang
• Is Part Of: Annals of nuclear energy. Volume 181(2023)
• Publication Date: 2023-02
• Subjects: Transient trend prediction -- Random forest model -- Equipment degradation -- Small modular reactor; Nuclear energy -- Periodicals; Nuclear engineering -- Periodicals; 621.4805
• DOI: 10.1016/j.anucene.2022.109507
• Languages: English
• ISSNs: 0306-4549

Abstract

Highlights: A transient trend prediction method based on random forest was proposed for SMR to predict safety parameters. The hyperparameters of the random forest models were optimized during training. Compared with a comparison model of backpropagation neural network, random forest model showed better prediction behavior. The proposed method can be utilized as an early warning tool, which enables the development of decision-making support technology for SMR, especially on the condition with unexpected equipment degradation. Abstract: Small modular reactor (SMR) has strong coupling characteristics and complex nonlinearities. Predicting the trend of SMR safety parameters and monitoring whether the parameter values exceed operational limits and conditions (OLC) before performing a transient helps the operator to judge the safety of the transient, especially in the case of equipment degradation. In this paper, a transient trend prediction method of SMR safety parameters based on random forest model was proposed, which considered the impact of equipment degradation on safety parameters. The transient data used for model training included different load-following conditions considering various degraded degrees of rotating speed drop in the main coolant pump (MCP) or effective heat transfer area reduction in the once-through steam generator (OTSG). To evaluate the model, the new untrained transient data was used as test data, the results showed that safety parameters were more accurately predicted by random forest, compared with the backpropagation neural network.