고속도로 2차 사고는 선행 사고(1차 사고) 또는 전방 고장 차량에 의해 교통흐름이 변화된 상황에서 발생하는 사고로, 이에 대한 효과적인 교통안전 관리전략이 필요하다. 그러나 일반사고에 비해 데이터 표본이 부족하여 신뢰성 있는 대응 전략 수립에 어려움이 있다. 본 연구는 고속도로에서 발생하는 2차 사고의 발생 주요 요인을 식별하고 예측하기 위해 BERT(Bidirectional Encoder Representations from Transformers) 기반 텍스트 분석 모델과 전통적 머신러닝 모델 (XGBoost, RandomForest, CatBoost)을 비교하였다. 교통사고 세부기록, 원클릭 속보자료 등 비정형 텍스트 및 정형 데 이터를 수집하고 1차 사고에 관한 시공간적 동적 변수를 통합하여 인공지능 기반의 사고 예측 프레임워크를 구축하였다. 특히, BERT 기반 모델을 통해 교통사고 문맥 정보를 고려하여 단어 삽입 및 대체 기법에 따른 2차사고 데이터 표본을 보완하였다. 또한, 설명가능한 AI(XAI) 기법을 활용하여 주요 사고 요인의 기여도를 시각적으로 해석하고 사고 예방 및 정책 수립에 필요한 정보를 제공하였다. 연구 결과, 제안된 하이브리드 접근법 기반 연구 프레임워크는 높은 정확도의 2 차 사고 발생 가능성 예측에 효과적이며, 교통사고관리시스템의 신뢰성과 효율성 향상에 핵심적인 기여를 할 것으로 기 대된다.

PURPOSES : This study focuses on advance traffic information to prevent secondary traffic accidents on express highways. The purpose of this study is to analysis the optimal scenario by evaluating the effect of each advance traffic information scenarios using virtual driving simulation. METHODS : By designing traffic information scenarios and services with a combination of VMS and mobile PUSH notifications, driver behavior in the event of a traffic accident was analyzed. For this, a simulation environment was designed through engineering analysis. Through virtual driving simulation, the driver's deceleration point and the perception-reaction time are analyzed. RESULTS : Four scenarios were designed and reviewed so that VMS and mobile PUSH notification can be provided simultaneously after the driver drove for 5 km. As a result of driving with 30 drivers, the drivers reacted fastest when VMS was installed, followed by mobile PUSH notification, VMS+mobile PUSH notification, and NOTHING. CONCLUSIONS : When designing traffic information service, it was observed that providing information through VMS alone is more efficient than providing two services of traffic information. Therefore, it can be used as basic data for preventing secondary accidents on express highway.

PURPOSES : This study aims to draw differences between primary and secondary crashes by comparing crash characteristics and to identify the unique characteristics of secondary crashes for making better effective countermeasures to reduce secondary crashes. METHODS : The characteristics of secondary crashes were compared to those of primary crashes through a two sample proportional test (z-test). RESULTS : The results showed that vehicle-to-vehicle crashes and vehicle-to-person crashes are dominant crash types in secondary crashes. Compared to primary crashes, secondary crashes were likely to occur during nighttime. With respect to season and weather, the proportion of secondary crashes occurred during winter and in snowy weather is relatively higher than that of primary crashes. The main causes of primary crashes were found to be drowsiness, speeding, and exaggerated steering control, whereas main factors affecting the occurrence of secondary crashes were negligence of keeping eyes forward and no keeping a safe distance as expected. CONCLUSIONS : The characteristics affecting the occurrence of secondary crashes are different from those of primary crashes, indicating that proper countermeasures should be established to prevent the occurrence of secondary crashes on highways.