원자력 발전소 사이버보안과 AI 융합: 기술 문서 활용 LLM 기반 CDA 유형 분류

초록

This research focuses on the classification of Critical Digital Assets (CDAs) based on NEI 13-10, a methodology for graded cybersecurity measures in nuclear power plants. It addresses the inefficiency and human error inherent in existing manual analysis methods by designing an automated classification model utilizing Large Language Models (LLMs). This study involves analyzing unstructured technical documents to refine the complex, original NEI 13-10 criteria tables, making them more suitable for LLM inference. Additionally, the study evaluates the model's performance by applying complex prompt engineering techniques, such as Chain-of-Thought (CoT), to control the model's reasoning process. Based on these experimental results, the research analyzes the feasibility of applying LLM-based automated models to regulatory classification tasks, discusses the limitations of text-based models, and proposes directions for future development.

키워드