Tsunami Risk Management Using Deep Learning and Probabilistic Methods

Abstract

Tsunami Risk Management Using Deep Learning and Probabilistic Methods Byung-Ho Kim Department of Civil and Environmental Engineering Graduate School of Hanyang University Advisor: Prof. Yong-Sik Cho This thesis is dedicated to exploring tsunami risk management at a nuclear power plant through the integration of deep learning and probabilistic methods. The primary objective is to underscore the critical necessity for precise and expeditious tsunami hazard predictions, while also aiming to generate safety analysis data through probabilistic tsunami hazard assessment (PTHA) by the Vine copula method. This dual-purpose approach is particularly essential for safeguarding coastal infrastructure, especially within the intricate domain of nuclear power plants. The Fukushima nuclear disaster and historical tsunami events serve as poignant reminders, emphasizing the urgency of formulating effective mitigation strategies for both long-term planning and real-time emergency response. For establishing tsunami risk management, this research adopts a multifaceted approach that includes the development of a one-dimensional convolutional neural network (CNN) model for generating predictive data. This innovative application enhances early warning systems and advances emergency response coordination, showcasing the model's efficacy in swiftly and accurately forecasting tsunami waveforms at the Uljin nuclear power plant (NPP). Additionally, the study conducts an exhaustive analysis of numerical simulation results, considering 1,107 cases initiated from various fault locations, to discern varying risks at nuclear power plants based on the epicenter's location. This analysis contributes to the development of tailored risk management strategies. Moreover, the research places emphasis on the development of probabilistic models, specifically leveraging the Vine Copula theory to formulate the Joint Probability Function. This innovative approach facilitates PTHA, estimating the exceedance probability for both maximum and minimum tsunami heights. These probabilistic models significantly may contribute to well-informed risk assessments and preparedness strategies. In conclusion, this study makes substantial contributions to the realm of tsunami risk management, providing valuable insights into enhancing safety and resilience for nuclear power plants. The integration of advanced modeling techniques ensures accurate predictions and informed decision-making in the face of complex geo-hazards, thereby enriching the broader discourse on coastal infrastructure safety in tsunami-prone regions. However, it is essential to acknowledge certain limitations inherent in this study, and these aspects are explicitly addressed in the Conclusions and Future Work sections.