Domain-Specific Large Language Model for Nuclear Cyber-Physical Scenario Generation

Cyber-physical systems are systems that integrate computational elements with physical processes via sensors or actuators to create the modern digital industrial system. Modern nuclear power plants (NPPs) require digital technologies, which turn most physical risks into cyber-physical risks. This paper presents cyber-physical attack scenarios for modern NPPs.The method utilizes historical nuclear incident data from the web databases of the International Atomic Energy Agency, the U.S. Nuclear Regulatory Commission, the Operational Performance Information System for NPPs of the Korea Institute of Nuclear Safety, the Common Weakness Enumeration, and the Common Vulnerabilities and Exposures to create a customized domain-specific large language model that analyzes patterns and security information to perform analysis on cyber-physical systems.The customized model can analyze large amounts of text data, identifying potential relations between cyber events and vulnerabilities and the impact these relations will have on system reliability and safety. Insights can be derived from the model through its generation of cyber-physical attack scenarios for NPPs, which are meant to increase the awareness of the safety and security needs of NPPs. The method of generating insights from publicly available databases not only contributes to the development of safe and secure NPPs, but can also benefit other fields.