Research on Nuclear Reactor Accident Diagnosis Method Based on Cross-Layer Collaborative Temporal Convolutional Network

To enhance the reliability and safety of nuclear reactors and improve the accuracy of reactor accident diagnosis, this paper establishes an intelligent accident diagnosis method based on reactor operational data. Taking the Qinshan 300MWe nuclear plant full-scope simulator as the research object, we propose a reactor accident diagnosis model based on a crosslayer collaborative temporal convolutional network (CL-TCN). The model extracts spatiotemporal features from reactor operational data at different window sizes using multiple TCNs with varying structural parameters. Subsequently, a random forest (RF) algorithm integrates the outputs of these networks to achieve cross-layer collaboration among the TCNs. We trained and tested the model using a simulator-generated dataset. The results demonstrate that the proposed model can accurately identify reactor accident scenarios and exhibits superior diagnostic performance compared to other deep learning methods in the control group. Therefore, the proposed approach provides a novel perspective for designing intelligent reactor accident diagnosis methodologies.