Prediction of multi-parameters of molten salt reactor based on deep learning networks

Molten salt reactor is one of the candidate reactor types for the fourth generation nuclear reactor. Accurate prediction of its key operating parameters can not only reveal the operating status of the reactor, but also provide early warning of abnormal working conditions, providing decision support and technical guidance for operators. Therefore, this article conducts predictive research on key parameters of molten salt reactors based on deep learning algorithms, in order to optimize the monitoring of the operating status of molten salt reactors and assist in the decision-making process. The RELAP5-TMSR program was used to establish a safety analysis model for the molten salt reactor system and generate a dataset. A multi parameter prediction model for the molten salt reactor was established based on deep learning methods such as autoencoder (AE), long short-term memory network (LSTM), and gated recurrent unit (GRU). The robustness of the model under noise was analyzed and optimized. The results show that the multi parameter prediction model for molten salt reactors based on the AE-GRU method has superior performance, with extremely high prediction accuracy and strong generalization ability. The average relative error of the predicted parameters is less than 0.04%, and it still exhibits high robustness in noisy environments, meeting the requirements of engineering applications. Therefore, the multi parameter prediction model for molten salt reactors based on the AE-GRU method can accurately predict the operating status of the reactor and provide reference for decision-making in molten salt reactor operation.