A transformer–XGBoost based model to fault diagnosis for CPR1000

With the rapid advancement of artificial intelligence (AI), intelligent diagnostics have seen broad application across industries. To address the limitations of traditional data-driven methods in accurately identifying faults in nuclear power reactor systems, this study proposes a hybrid model combining Transformer and XGBoost for diagnosing faults in CPR1000 pressurized water reactors. Fault-related data were automatically collected using the self-developed AutoSave-PCTRAN software from the PCTRAN simulator, with key nuclear parameters selected as features. A 10-fold cross-validation with recursive feature elimination was used for feature selection. The Transformer model extracted temporal features via its self-attention mechanism, and an RWOA was employed to tune XGBoost hyperparameters for fault classification. The model effectively identified faults such as Loss of Coolant Accident(LOCA), Steam Line Break Inside Containment(SLBLC), and steam generator B-tube rupture(SGTR-B), achieving 99.46% accuracy, confirming its reliability and practicality.