[2M15-21] Large Scale Simulation

[2M21] Development of LSTM models for the combined forecasting and diagnosis of PWR LOCA accident events

*Johndel Baluyot Obra¹, Shuichiro Miwa¹ (1. UTokyo)

Keywords:Nuclear safety, Real-time diagnostics, Deep learning, LSTM, LOCA

The real-time diagnosis of loss of coolant accidents (LOCA) in a pressurized water reactor (PWR) by forecasting the progress of the accident and predicting its cause is important in aiding plant operators with their decision-making process and accident response strategy. In a previous work, the potential of Long Short-Term Memory (LSTM) models in accident diagnosis was established by predicting the extent of pipe damage and location of pipe break from simulated LOCA events that were generated from RELAP5/SCDAPSIM. In this study, a forecasting model was supplemented to the existing diagnostic model to develop an LSTM-based real-time LOCA accident forecasting and diagnosis tool. The combined forecasting and diagnosis model has better capabilities and less real-time prediction error compared to the diagnostic model alone.