Abstract
Predicting conflict risk from a microscopic perspective is crucial for understanding traffic dynamics and preventing crashes. However, previous studies often missed this perspective and lacked methods for applying it in spatial contexts. This study introduces a method that predicts conflict risk using both spatial and temporal segments, using real-time data for practical implications. Employing a hybrid model that combines long short-term memory (LSTM) and a convolutional neural network (CNN), the research utilizes three hours of vehicle trajectory data from the Shanghai Inner Ring Expressway. Specifically, a 10 s data window is employed to forecast conflict risk within the subsequent 3 s, 4 s, and 5 s intervals, considering a 10 m spatial section. Modified time to collision (MTTCL) is adopted to assess conflicts between vehicles. The study incorporates 12 factors to predict conflict risk, prioritizing them based on SHAP values. The results highlight that most conflicts occur within the 3 s prediction horizon, with prediction accuracy decreasing as the forecast horizon extends. Notably, the study identifies key contributing factors, including the speed difference between vehicles, the number of lane changes, inter-vehicle distance, distance variability, and the number of cars. Spatially, the research identifies entry and exit ramps and their surrounding areas on the expressway as hotspots for sideswipe conflicts.
Disclosure statement
No potential conflict of interest was reported by the authors.
Data availability statement
The data described in the methodology section is available upon request for noncommercial purposes with proper reasons. Data request link: https://magic.tongji.edu.cn/kycg/MAGICsjj.htm.