Development and validation of machine learning-based transient identification models in a liquid-fueled molten salt reactor system

Safety is the most important aspect of nuclear power plants. Rapid identification and effective prevention of accidents in nuclear reactor system is a significant method to enhance the safety of the current fleet of reactors. Machine learning (ML) has been introduced in engineering applications of nuclear power plants and is becoming increasingly practical and powerful in recent years. Consequently, ML can also benefit rapid transient identifi-cation in nuclear power plants. The feasibility of ML-based identification models to identify transient events in liquid-fueled Molten Salt Reactor (MSR) is presented. Four transient identification models based on recurrent neural network (RNN), support vector machine (SVM), decision tree (DT) and k-nearest neighbor (KNN) were developed and validated. RELAP5-TMSR code was used to generate datasets including eleven operation condi-tions, and these datasets were used to train, optimize, and validate the identification models. Four metrics including accuracy, precision, recall and F1 score were utilized to evaluate all four identification models. Moreover, the robustness of the models under noise was tested. The four ML-based models were successfully applied to transient identification of liquid-fueled MSRs. The KNN-based model has the best performance and achieves high test scores under noise. In the future, these proposed intelligent identification models will have good potential and prospects in supporting the operation of nuclear power plants.