In Loss of Coolant Accident (LOCA) scenarios, the reactor core temperature may rise sharply, potentially causing fuel damage and radioactive material release. This paper proposes a method combining Latin Hypercube Sampling (LHS) and Physics-Informed Neural Networks (PINNs) for simulating and solving LOCA conditions in nuclear reactor accidents. By generating high-quality parameter space training samples using LHS and training the PINNs model under physical constraints, the method effectively simulates the complex fluid behavior under LOCA conditions. Evaluation results of the model show an average absolute error (MAE) of 0.054, a root mean square error (RMSE) of 0.132, and a coefficient of determination (R²) of 0.85, validating the effectiveness of the proposed method in complex conditions.

nuclear reactor accidents, LOCA, LHS, PINNs