딥러닝을 이용한 중대사고 상황 시 격납건물 내 수소농도 예측

Abstract

In the severe accidents in nuclear power plants (NPPs), the hydrogen combustion phenomenon occurs by oxidation of the fuel cladding and molten corium-concrete Interaction (MCCI) causes damage to the equipment. More than 4% of the hydrogen in the containment is considered to be a threat to the integrity of the NPPs. Besides, it is difficult for operators to make accurate judgments and actions, because major safety parameters that need to be monitored in NPPs change very quickly in the early stages of accidents. In order to effectively manage severe accidents, it is essential to identify key variables for a very short time leading up to the accidents. Therefore, the purpose of this paper is to accurately predict the concentration of hydrogen in containment buildings to prevent further accidents and to manage the incident more efficiently. In this thesis, a deep learning model of data-based artificial intelligence(AI) methods is developed to predict the hydrogen concentration inside containment after a reactor trip in an event of severe accidents. The Deep Neural Network (DNN) algorithm used as a predictive model is trained and verified using a large amount of data obtained by the simulation of optimized power reactor 1000 (OPR1000). Because applying the accident data of actual NPPs is limited, the data were acquired using the Modular Accidents Analysis Program (MAAP). The loss of coolant accident (LOCA) situations were simulated and various data were obtained by considering the behavior of the passive auto-catalytic recombiner (PAR), which affects the concentration of hydrogen in containment buildings. There have been several studies to predict the hydrogen concentration by using fuzzy neural network (FNN) and cascaded fuzzy neural network (CFNN). Through the comparison of their models and the proposed DNN model, the accuracy of DNN model is verified.