Automated anomaly response in nuclear power plants to enhance nuclear safety

Dong, F.*; 出町 和之*; 吉川 雅紀 ; 高屋 茂  

Dong, F.*; Demachi, Kazuyuki*; Yoshikawa, Masanori; Takaya, Shigeru

To enhance nuclear safety, condition-based maintenance enables real-time monitoring and supports anomaly detection and emergency management. This research applies artificial intelligence to provide promising solutions by facilitating automated anomaly response procedures and reducing human errors, which integrated a deep reinforcement learning framework with a surrogate model to replace computationally expensive nuclear power plants simulation codes. The proposed framework reduces computational costs, accelerates decision-making, and provides a long-term perspective on system behavior after each action.