Earticle

초록

영어

This study proposes the H.E.L.P (Human Evacuation Location Prediction) model, which reconstructs National Fire Data System (NFDS) data using artificial intelligence to quantitatively predict evacuee behavior patterns and rescue priorities during fires. Despite extensive NFDS records, current use is confined to statistical and administrative reporting, with limited support for real-time decision-making. An AI training dataset integrating individual characteristics (age, gender, disability, movement type) and environmental factors (fire location, smoke spread, emergency stairway accessibility) was developed and applied to deep learning and SARSA-based reinforcement learning simulations. Experimental results demonstrated 96–100% recognition accuracy, over 80% prediction accuracy, and more than 20% reduction in average evacuation time. Additional tests using RandomForest and XGBoost achieved high casualty prediction accuracy, confirming NFDS’s potential for real-time rescue strategy support. The model’s integration with IoT-based recognition and AR/VR command systems can enhance precision and efficiency in firefighting operations, offering a scalable public safety AI platform that unites policy, technology, and data.

목차

Abstract
1. 서론
1.1 연구의 배경
1.2 연구 목적 및 의의
1.3 용어의 정의
2. 연구의 방법 및 범위
2.1 연구의 방법
2.2 연구 흐름도
2.3 연구의 범위
3. 이론적 배경
3.1 현실태 및 문제점
3.2 선행연구의 고찰
3.3 운영사례
3.4 시사점 도출
4. ‘화재현장’ 인간 행동패턴 예측 모델(H.E.L.P: Human Evacuation Location Prediction13)) 개발 (제안)
4.1 AI 딥러닝을 위한 사전학습 자료 구성 (시각적 특징: 사람 인식Ⅰ)
4.2 AI 딥러닝을 위한 사전학습 자료 구성(시각적 특징: 사람 인식Ⅱ)
4.3 AI 딥러닝을 위한 사전학습 자료 구성(방향성 인식)
4.4 AI 딥러닝을 위한 강화학습 자료 구성 / OCR 인식률
4.5 화재현장 ‘인간 행동패턴 예측 모델(H.E.L.P)’의 개발
4.6 H.E.L.P 모델 개발의 소결
5. 실증연구
5.1 인공지능(AI) 기반 행동패턴 모델 연구
5.2 실증연구의 소결
6. 결론
7. 정책제언
7.1 NFDS의 예측·대응 지원 플랫폼 고도화 추진
7.2 NFDS 데이터 성능 저하 요인 분석 및 보완 절차 개선 (연구의 한계 / 향후 연구과제)
참고문헌

키워드

저자

• 윤홍범 [ Hong-Bum Yoon | 대전서부소방서 ]
• 고택훈 [ Taek-Hoon Go | 대전서부소방서 ]
• 최웅현 [ Woong-Hyun Choi | 대전서부소방서 ]
• 진성덕 [ Sung-Deok Jin | 대전소방본부 ]

참고문헌

발행기관

간행물

표지보기 관심저널 등록

• 간행물명

  한국화재감식학회 학회지 [Magazine of Fire Investigation Socity of Korea]

• 간기

  계간

• pISSN

  2092-531X

• 수록기간

  2009~2026

• 등재여부

  KCI 등재

• 십진분류

  KDC 539 DDC 628