Comparative Study of Three Machine Learning Algorithms for Reactor Operation Parameter Inference

Abstract This study presents a data-driven approach for predicting nuclear reactor operation state parameters using machine learning algorithms. We developed a multi-input single-output (MISO) framework to model key parameters such as temperature, boron concentration, R-rod position, burnup, and power level, monitored by the digital control system (DCS). Using extensive datasets from a certain nuclear reactor's unit 5/6, we evaluated the performance of K-nearest neighbors (KNN), random forest (RF), and extreme gradient boosting (XGBoost) algorithms. The predictive accuracy was significantly improved through meticulous parameter tuning and the introduction of a comprehensive set of evaluation metrics, including mean absolute error (MAE), mean square error (MSE), and mean relative error (MRE). RF and XGBoost outperformed KNN, indicating the superiority of ensemble models. The models also showed strong generalization capabilities when tested on a separate dataset. The resulting regression model offers an effective tool for accurate prediction of reactor state parameters, enhancing the safety and efficiency of nuclear power plant operations.