Development of a novel data-driven NPP simulation method based on physics informed neural network for dynamic PSA application

Abstract

Because nuclear power plants (NPPs) are safety-critical systems with large sizes and high complexities, various methods have been developed to identify possible accidents and address potential risks. However, the current PSA has two limitations. The first is a data dependency issue, and the other is a dynamic interaction issue. The dynamic interaction can be classified into dynamic interaction with long-time constants and short-time constants. The dynamic interaction with a long-time constant can be considered in prognostics. However, the dynamic interaction with a short-time constant is hard to consider in the current stage because of the calculation speed of the conventional code analysis model. In addition, quantitative analysis of the interaction (ex., actions from safety systems) that affects the state variable is difficult. Therefore, in this study, we proposed a physics-informed neural network-based data-driven simulation framework to consider the calculation speed and interaction issues. The effectiveness of the methodology is confirmed by comparing the proposed model with the conventional analysis method. The proposed study is expected to contribute not only to the dynamic probabilistic safety assessment but designing a digital twin or creating a new simulator.