Using ex-core detectors and deep neural networks for monitoring power distribution in small space reactors

Abstract

Ex-core detectors have the potential to monitor the power distribution in small space reactors. However, there are still considerable challenges remain in their practical implementation. To address this gap, this paper proposes a novel method to monitor power distribution utilizing ex-core detectors and deep neural networks. A small space reactor model simplified from TOPAZ-II was constructed based on the assumption that 12 ex-core detectors could be applied. New neural network models were established to consider the differences of pin power at different positions in the core by independently modeling the inner and outer fuel pins. This method was extensively validated across a wide range of operational conditions. The deep neural network method also exhibits reduced sensitivity to noise. By training on datasets containing noisy signals, the neural network method can handle signals containing ± 1 % noise while the accuracy of power distribution predictions is maintained. In addition, the deep neural network method is capable of monitoring asymmetric power distribution. By learning the characteristics of signals from asymmetric detectors, this method can accurately predict core power distribution even under abnormal operational conditions.