Real time estimation of radionuclides in the receiving water of an inland nuclear power plant based on difference gated neural network. (November 2020)

• Record Type: Journal Article
• Authors: Zhang, Yi-Jing; Hu, Li-Sheng
• Is Part Of: Radiation physics and chemistry. Volume 176(2020:Nov.)
• Publication Date: 2020-11
• Subjects: Radionuclide estimation -- Difference gated neural network -- Radionuclide concentration -- Nuclear power plant; Radiation chemistry -- Periodicals; Radiometry -- Periodicals; Radiation -- Periodicals; Chimie sous rayonnement -- Périodiques; 539.2
• DOI: 10.1016/j.radphyschem.2020.109019
• Languages: English
• ISSNs: 0969-806X

Abstract

Abstract: Estimation of radionuclides in closed water is an essential licensing issue since it concerns the surroundings' safety, especially for inland sites. As the receiving water of inland nuclear power plant, closed water's flow processes are slow, unstable as well as highly nonlinear, leading to difficulty in monitoring of radionuclides. The traditional method aims to detect the photopeak of γ spectrum, which is complicated and severely time-consuming for the sampling and peak detection processes. Motivated by its time-consuming computation, a real time method for radionuclide estimation in closed water is proposed. A self-defined difference gated neural network utilizing radionuclide concentration data generated from Environmental Fluid Dynamics Code to perform estimation is mainly adopted. Simulation results show that compared to multilayer perceptron and long short-term memory models, the proposed difference gated neural network achieves the accuracy of 98.7% when estimating H 3, C 60 o, A 110 m g and C 58 o . Accordingly, the designed method can be used for monitoring of nuclear power plant's waste water, especially in closed waters. Highlights: Difference gated neural network (DGNN) identify nuclide types in water perfectly. DGNN is slower than multilayer perceptron, but with giant accuracy improvement. DGNN is superior to long short-term memory in terms of accuracy and training time. The accuracy of DGNN depends mainly on the learning rate and hidden dimensions. DGNN is competent in characterizing the dynamics and nonlinearity of time sequence.