Enhanced accident diagnosis for molten salt reactors using an adaptive residual CNN with Bayesian optimization

摘要: Safety is of paramount importance in nuclear power plants. Accurate and reliable accident diagnosis is essen#2;
tial for ensuring operational safety in reactor systems. The convergence of Industry 4.0 technologies and deep
learning methods has emerged as a promising approach for improving the operational safety of nuclear energy
systems, particularly in fault detection and diagnosis (FDD) applications. This study proposes a novel adaptive
accident diagnosis framework tailored for molten salt reactors (MSRs) based on an enhanced residual convo#2;
lutional neural network (AM-RCNN). The AM-RCNN incorporates an anti-noise module implemented using
the Soft Threshold method, together with an attention mechanism, to improve robustness. Datasets representing
eight distinct operational scenarios were generated using the RELAP5-TMSR simulation tool. An appropriate
subset of input features for MSR accident diagnosis was selected using Pearson correlation analysis and random
forest importance ranking. The models were subsequently trained, validated, optimized, and tested. Compar#2;
ative analyses with conventional RCNN and CNN architectures demonstrate the diagnostic advantages of the
proposed approach. In addition, the integration of Bayesian optimization further enhances the performance of
the AM-RCNN. As a contribution to intelligent monitoring research for MSRs, the proposed method provides
reliable decision support for nuclear system operation, particularly in autonomous scenarios.