摘要详情

ID / 提交时间

51 / 2024-06-13 21:09:19

标题

Research on CHF look-up table prediction based on deep learning

关键字

CHF,One-dimensional convolutional neural network,BP neural network,Algorithm-BP neural network,Deep learning

主题及专题

TRACK 5: NUCLEAR ENGINEERING

状态

终稿

作者

Kai Wu / China Nuclear Power Technology Research institute

Lei Zhang / China Nuclear Power Technology Research institute

Xiong Guo Liu / China Nuclear Power Technology Research institute

lu jiaxin / Harbin Engineering University

Hetao Sun / Harbin Engineering University

摘要

Abstract: In nuclear fuel rod bundles and steam generators, the heating surface may fail once the wall surface exceeds the CHF, the mechanism of CHF generation is complex, and there is no single theory or equation that can be applied to all CHF conditions. The CHF lookup table is basically a normalized database that predicts CHF as a function of coolant pressure, mass flux, and thermodynamic mass. The deep learning model can effectively predict CHF, and its prediction performance is simpler and more accurate than traditional relational fitting. Based on the data of the CHF lookup table, we use BP neural network, genetic algorithm BP neural network, and one-dimensional convolutional neural network to predict the predicted value of CHF within the measurement range, and compares it with several common empirical relationships. The results show that the prediction performance of the three deep learning models is improved to varying degrees compared with the empirical relationship. BP neural network is the fastest to train; The genetic algorithm -BP neural network has the best prediction performance, with MAE=1.490%, MSE=0.03%, and R²=0.976, but the training time is too long and requires a lot of resources. The prediction performance of the one-dimensional convolutional neural network is slightly lower than that of the GA-BP neural network, with MAE=1.574%, MSE=0.06%, and R²=0.968, and the training time is comparable to that of the BP neural network, which occupies less resources. The research results provide a new way for the prediction research of CHF look-up table.