Application of multi-input multi-output prediction model to reactor accident parameters prediction

In order to meet the requirement of multiple target parameters prediction under reactor accident, a multi-input multi-output prediction model based on recurrent neural network is proposed. Based on the data samples from protected transient overpower (PTOP) accidents of lead-bismuth fast reactor NCLFR-Oil, the mass flow rate, core outlet temperature and maximum temperature of cladding were predicted by multi-input multi-output model coupled multivariable optimization algorithm. The prediction effects of gated recurrent unit (GRU) and long short-term memory (LSTM) on different variables are compared and analyzed from the aspects of prediction accuracy and prediction time. The results show that the prediction effect of the model is acceptable under multi-target conditions, in which the MAPE of the core outlet temperature is up to 0.162%, and the RMSE is up to 0.798℃. What's more, GRU can predict the core outlet temperature and the maximum temperature of the cladding better, while LSTM can predict the mass flow rate more accurately. The multi-input multi-output parameter prediction model in this paper has more practical significance for engineering application.