In this study, we employ a Gated Recurrent Unit (GRU) neural network model to predict changes in coolant flow in small-scale nuclear power plants under Loss-of-Coolant Accident (LOCA) conditions with high precision and speed. We quantitatively compare this model with the traditional Long Short-Term Memory (LSTM) method to validate the accuracy of the GRU model. Additionally, we perform an interpretability analysis on the predictive model using the SHAP (SHapley Additive exPlanations) method for the first time, revealing the inherent influencing factors in $L O C A$ accidents. This research aims to enhance trust in artificial intelligence models in nuclear power scenarios, facilitate a better understanding of prediction results and the current system state, and promote the practical application of artificial intelligence models in nuclear power systems.