[2G14-21] AI Technology

[2G20] Integration of AI Technology and Thermal Hydraulics for the Development of a Data-Driven Methodology for Plant Safety Assessment

(5) Investigation of LSTM models for the prediction of LOCA accident events

*Johndel B Obra¹, Shuichiro Miwa¹ (1. The University of Tokyo)

Keywords:Nuclear safety, Deep learning, LSTM, LOCA

The loss of coolant accident (LOCA) in a nuclear power plant is crucial to monitor because it tends to escalate to failure modes that can result in severe structural damage. This study investigates the potential of machine learning models in diagnosing loss of coolant accidents. Specifically, the potential of Long Short-Term Memory (LSTM) models using the time series data of plant variables during LOCA events generated from the RELAP5/SCDAPSIM was investigated. The results showed that LSTM models are capable of classifying the location and predicting the extent of pipe damage. Furthermore, the contribution of each plant variable to the predictions was determined using their Shapley values which then imply their importance to the LSTM model.