A machine learning diagnosis of the severe accident progression

Abstract

A Machine Learning (ML) platform is proposed to assist the operator in diagnosing the severe accident progression, where some of the signal is corrupted, and/or estimation of key parameters for accident management is needed. To predict the key parameters for accident management such as lost signals for the pressure and core water level, the amount of hydrogen generation, and the mass of nuclear aerosols, a Machine Learning (ML) platform with long short term memory (LSTM) network is proposed, where multiple accident scenarios can be accumulated and used for training. Training and test data were produced by MELCOR simulations of the Fukushima Daiichi Nuclear Power Plant (FDNPP) accident at unit 3. The MELCOR simulation data is presented as time series at equal intervals. Feature variables were selected among many plant parameters, where the importance ranking was determined by a recursive feature elimination technique using RandomForestRegressor. We performed a sensitivity study of the different choices of training and test scenarios, the number of feature variables, the number of neurons, and target variables on the prediction accuracy of the ML platform. Among eight MELCOR simulation results, different combinations of training and test data are selected in terms of similarity between them. It is found that when the number of feature variables is more than five, the proposed ML platform was able to predict not only similar test data but also unseen test data with reasonable accuracy. It is also found that the proposed ML platform consistently predicts any of the lost signals and key parameters for accident management with reasonable accuracy. Pre-trained ML model predicted the target variable within several minutes on a desktop computer.