A Fault Diagnosis Model for Multi-condition Reactor Coolant Systems Based on the SSA-Elman Neural Network

The Reactor Coolant System (RCS) is the core safety barrier within the nuclear island of a nuclear power plant, and the rapid identification of its multi-condition operational states is vital for ensuring safe reactor operation. This paper proposes an intelligent fault diagnosis model based on the Elman neural network optimized by the Sparrow Search Algorithm (SSA), referred to as the SSA-Elman model, for multi-state fault identification in RCS. A simulated dataset encompassing eight representative operating conditions-normal operation, small/large break loss-of-coolant accidents (LOCA), steam generator tube rupture (SGTR), and main pump flow loss (S1-S8)-was constructed using the MARS-KS simulation platform. Each condition includes 500 labeled samples covering 83 critical operational parameters. The dataset was then divided into training and validation sets, and both Elman and SSA-Elman models were trained to evaluate their predictive performance under various fault conditions.Experimental results demonstrate that the SSA-Elman model significantly outperforms the traditional Elman network across all three evaluation metrics-mean squared error (MSE), root mean squared error (RMSE), and mean absolute error (MAE). In the training set, the maximum error reduction reached 88.2%, with substantial improvement also observed in the validation set. These results highlight the model’s superior nonlinear modeling capability and generalization performance. This study confirms that the SSA-Elman model can effectively improve fault identification accuracy in RCS under uncertain conditions and provides a valuable intelligent tool for supporting the safe and reliable operation of nuclear power systems.