[2C01-06] Development of Evaluation Methods, Safety Requirements for New NPPs

[2C02] Framework for Intelligent Generation of Countermeasures against Emergencies at Nuclear Power Plants

*Satoshi Abe¹, Kazuyuki Demachi¹, Shi Chen¹ (1. UTokyo)

Keywords:reinforcement learning, surrogate AI model, countermeasures against accidents

In order to ensure nuclear security and safety in nuclear facilities, we need to take countermeasures against beyond design basis events (BDBE). In addition, the nuclear industry is lagging in digitalization, and we need to work on the implementation of digitalization. In this study, I focused on nuclear safety and aimed to construct a framework to automatically generate countermeasures for beyond design basis accidents (BDBA) using deep learning. I employed reinforcement learning, in which the agent selects the optimal action through trial and error of various actions in the environment. In doing so, a surrogate AI model of a plant simulator was constructed and used in order to reduce the time required for reinforcement learning. In this time, I constructed the above framework and verified whether it was able to propose suitable countermeasures against accidents.