Development and implementation of a surrogate model for bare critical Mass estimation in nuclear fuel cycle simulator

In this study, estimation functions of bare critical mass (BCM) and heat content (HC) of nuclear materials were developed and introduced into a nuclear fuel cycle simulator, the Nuclear Material Balance (NMB), in order to use them for non-proliferation studies of future advanced nuclear fuel cycles. HC is defined as the total decay heat of one BCM of material. To reduce the calculation cost of Monte Carlo calculation for BCM, a surrogate model was developed. Several regression models were compared using BCM data calculated by MCNP-6.2 and JENDL-5 for randomly generated actinide compositions, and a multi-layer perceptron (MLP) was selected as the most appropriate model. The optimized MLP model had two hidden layers with 38 neurons in each layer, and the 99% confidence interval of the relative error was ±5.2%. The model was implemented into NMB, and time changes of BCM and HC were calculated for five types of plutonium separated in light water reactor and fast reactor reprocessing scenarios. For five representative plutonium compositions, additional MCNP calculations were performed, and agreement with the NMB calculation was confirmed within a 1.6% error. These results show that the developed MLP surrogate model can estimate BCM and HC with sufficient accuracy and small calculation cost in nuclear fuel cycle simulation.