Time series usually have the characteristics of linear and nonlinear. Single model has certain limitations, which proposes mixed models for time series prediction.

This study aims to explore the application of the combination of Mann-Kendall test, differential autoregressive mobile average model (ARIMA) and long and short-term memory (LSTM) used in the prediction of the number of operational events of nuclear power plants (NPP) collected in Nuclear Safety of China and Annual Report of Nuclear Safety.

Firstly, the R software was used to build ARIMA (2,1,2) model to obtain the linear part of operation events with the number of nuclear power plant operation events from 1991 to 2018, and LSTM model was developed to predict the deviation sequences, hence the nonlinear part of the number of operation events was derived from those deviation sequences. Then, combined model of ARIMA and LSTM was established to predict the number of operational events. Finally, the predicted values based on measured data were verified by actual measured data.

The verification results show that the ARIMA and LSTM combination model can be employed to improve the prediction accuracy effectively by 3%, and the predicted values of operation events in nuclear power plants from 2019 to 2020 are similar to the data collected in Annual Report of Nuclear Safety.

The combined model can better fit the time series of the number of operating events of NPP and correct the error of the single model.