Stochastic estimation of radionuclide composition in wastes generated at Fukushima Daiichi nuclear power station using Bayesian inference

Abstract

ABSTRACT A modeling calculation methodology for estimating the radionuclide composition in the wastes generated at the Fukushima Daiichi nuclear power station has been upgraded by introducing an approach using Bayesian inference. The developed stochastic method describes the credible interval of the regression curve for the log-normal distribution of the measured transport ratio, which is used to calibrate the radionuclide transport parameters included in the modeling calculation. Consequently, the method can predict the probability distribution of the radionuclide composition in the Fukushima Daiichi wastes. The notable feature of the developed method is that it can explicitly investigate the improvement in the accuracy and confidence (degree of belief) of the estimation of the waste inventory using Bayesian inference. Specifically, the developed method can update and improve the degree of belief of the estimation of the radionuclide composition by visualizing the reduction in the width of uncertainty in the radionuclide transport parameters in the modeling calculation in accordance with the accumulation of analytically measured data. Further investigation is expected to improve the credibility of waste inventory estimation through iteration between modeling calculations and analytical measurements and to reduce excessive conservativeness in the estimated waste inventory dataset.