Studying merge behaviour in weaving segments: insights from traffic conflict prediction and risk factors analysis

Le résumé fourni par la source

Within the weaving area, merging vehicles are required to complete lane-changing (LC) maneuvers within a limited distance, potentially increasing collision risk. Under specific driving intentions and road infrastructure, merging behaviours may exhibit recognisable cluster patterns. This paper aims to develop a real-time traffic conflict prediction model based on microscopic vehicle trajectories and to identify the risk factors contributing to dangerous merge behaviours, by drawing insights from LC patterns. To capture variations in merging behaviour, four clustering algorithms are used to extract LC patterns. Furthermore, a CNN-LSTM model for traffic conflict prediction is developed by combining the feature extraction capability of Convolutional Neural Network (CNN) with the gated memory mechanism of Long Short-Term Memory (LSTM). Additionally, three random parameter logit (RP-logit) models are used to analyze risk factors associated with dangerous merging behaviours using pre-conflict data. The results benefit the development of individualised ADAS intervention strategies to prevent unsafe merging behaviour.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.