Explainable Artificial Intelligence Based Classification of Normal and LOCA Conditions in APR-1400 Using GPWR Safety Parameters

Abstract Loss of Coolant Accident (LOCA) remains one of the most important transient conditions considered in nuclear reactor safety analysis. Early identification of LOCA type and operating condition is necessary for safe plant response and monitoring. In this study, machine learning was applied to classify 13 operating conditions of the APR-1400 reactor. Normal operation together with small, medium, and large LOCA events in both hot-leg and cold-leg locations were investigated. The dataset was generated using the Generic Pressurized Water Reactor (GPWR) simulator. Thermal-hydraulic and neutronic parameters under transient conditions were also considered. A feedforward neural network was trained using the generated data. After that the data was evaluated using classification accuracy, confusion matrix analysis, and F1-score. The model reached an overall classification accuracy of 97.9% on 7,000 test samples. Most operating conditions were identified correctly, while limited overlap was observed between a few similar LOCA cases during the early transient stage. Explainable AI is used to improve model output. SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-agnostic Explanations) were used as explainable AI methods. The results showed that the pressure and coolant temperature had the strongest effect on the LOCA scenarios. These trends obtained agree with the expected reactor behavior during abnormal conditions. The results indicate that the proposed framework can support reactor condition monitoring and LOCA classification in APR-1400 systems and may be extended in future work using additional plant operating data.