Development of accelerated prediction method using artificial neural network for nuclear power plant severe accident application

Abstract

During a severe accident in a nuclear power plant, since there is a great deal of uncertainty about the progression of a severe accident, it is necessary to check the condition of the nuclear power plant and determine the available mitigation strategies by following the Severe Accident Management Guidelines (SAMG). This is a very stressful situation and human error can make accident consequences worse. To resolve this issue, the importance of the Accident Management Support Tool (AMST), which helps make decisions under stressful accident situations, has emerged. AMST can be divided into a part that predicts an accident and a part that makes a decision through repeated calculation for various scenarios. Changes in thermohydraulic/physical parameters due to a selection of strategies should be computable in fast manner. To this end, a model that can make predictions quickly by applying a neural network is developed in this thesis. By using MAAP 5.03 severe accident calculation code, the change of thermohydraulic/physical variables was calculated under various component failure/mitigation strategies were performed for the purpose of data generation. A regression model was created for the generated data with a neural network. R2 value of the model was 0.98, which was satisfactory, and it confirms that the regression was successful. As a result of comparing the time series data that come from repeating 72 times calculations of the model based on neural network to predict the MAAP data, the error is small on average for all scenarios, but the trend in certain variables was predicted in opposite directions which indicates further improvement in the model is necessary in the future.