Multivariate Time Series Prediction for Loss of Coolant Accidents With a Zigmoid-Based LSTM

Post-LOCA prediction is of safety significance to NPP, but requires a processing coverage non-linearity, both short and long-term memory, multiple system parameters. To enable an ability promotion previous LOCA models, new gate function called zigmoid introduced embedded the traditional long short-term memory (LSTM) model. The newly constructed zigmoid-based LSTM (zLSTM) amplifies gradient at far end time series, which enhances without weakening one. Multiple parameters are integrated into 12-dimension input vector zLSTM for comprehensive consideration based on can be accurately generated. Experimental results show accuracy evaluations progression produced by proposed zLSTM, two baseline methods demonstrating superiority applying LCOA predictions.