Inversion Method for Multiple Nuclide Source Terms in Nuclear Accidents Based on Deep Learning Fusion Model

During severe nuclear accidents, radioactive materials are expected to be released into the atmosphere. Estimating source term plays a significant role in assessing consequences of an accident assist actioning proper emergency response. However, it is difficult obtain information on directly through instruments reactor because unpredictable conditions induced by accident. In this study, deep learning-based method estimate with field environmental monitoring data, which utilizes bagging fuse models based temporal convolutional network (TCN) and two-dimensional neural (2D-CNN), was developed. To reduce complexity model, particle swarm optimization algorithm used optimize parameters fusion model. Seven typical radionuclides (Kr-88, I-131, Te-132, Xe-133, Cs-137, Ba-140, Ce-144) were set as mixed terms, International Radiological Assessment System generate model training data. The results indicated that average prediction error for seven nuclides test less than 10%, significantly improved estimation accuracy compared obtained TCN or 2D-CNN. Noise analysis revealed robust, having potential applicability toward more complex scenarios.