Prediction of core unmeasurable parameters during loss of coolant accident using deep neural network method

The parameters within the reactor core would be categorized into two different groups of measurable and unmeasurable parameters. The determination of unmeasurable parameters, such as void fraction and critical heat flux, plays a significant and fundamental role in predicting the occurrence of accidents and emergency situations within the reactor. The utilization of deep neural networks represents one of the methods for accurate and reliable estimation of these parameters. Such estimations facilitate the implementation of necessary measures to prevent accidents or mitigate their consequences. In this study, three deep neural network models namely LSTM, TFT, and NBEATS are employed for void fraction prediction within the reactor core after Loss of Coolant Accident (LOCA). The neural network training will be performed without covariates, using past covariates and using future covariates. The results reveal that the TFT neural network, trained with future covariates (e.g. pressure, temperature, water velocity and steam velocity) yields the lowest error.