방사성물질의 방사선원항 추정을 위한 인공지능 및 드론 적용에 관한 연구

Abstract

The reliable prediction of air concentrations due to the atmospheric release of radioactivematerials under accidents at nuclear facilities is greatly valuable in guiding an effective and timelyresponse. Those predictions start with quantifying the radiation source terms released to the environmentin nuclear facilities, but a difficulty in their quantification is uncertainties of the results obtainedfrom drones are contained in atmospheric dispersion models and input variables consisting the models. A try to apply the artificial neural network for a quantitative prediction of the radiation source termsis introduced for an acute release under an accident in nuclear facilities in this paper. Deep neuralnetwork (DNN) model is employed for predicting them and two dimensional simulation by advectiondispersion equation using the upwind scheme, of which is a numerical method based on Euleriansystem, is introduced. To testify the capability of predicting source terms, the machine learning byDNN model is then applied to decide the points capable of predicting source terms and radioactiveconcentrations released to the atmosphere with the availability of predictions. The accuracy metrics ofR2 score for regression and F1 score for classification are to be 0.99, respectively and the consumedmemory by DNN model is reduced about 1,800 times compared to the results using database from theacute release scenario. Score solutions by DNN are driven to the result that DNN model can be fast andaccurate in predicting the source terms with the information of meteorological data and contaminateddata. It is concluded the prediction of radioactive materials driven to the atmospheric dispersion modeland the artificial neural network can be faster and more reliable one of source terms released to theenvironment.