Thermal Power Prediction of Nuclear Reactor Core based on LSTM

Nuclear reactors often operate at varying loads. In order to match the core thermal power with the load while ensuring the safe operation of the nuclear power plant, we estimate the core thermal power value of the nuclear reactor for timely adjustment. The traditional physical models and experimental methods based on nuclear reaction mechanism cannot get the accurate value of thermal power by directly analyzing the relevant data. We construct a long short-term memory network (LSTM) by using the real monitoring data of a CANDU reactor nuclear power plant. Based on the self-learning characteristics, the LSTM can precisely predict the core thermal power of nuclear reactor. Simulation results show that the proportion of data with absolute error less than 50MW is 97.63%. The proportion of data with relative error less than 5% is 98.73%, and the average relative error is 2.65%. Furthermore, we compared the performance of LSTM with the back-propagation neural network (BPNN) and the convolution neural network (CNN), where the LSTM model outperforms the BPNN and CNN in terms of prediction precisions.