关键词: 智能交通, 风险预测, 驾驶行为熵, 辅助驾驶

Abstract: As the critical nodes in urban transportation networks, the accident rates of intersections remain persistently high. The development of assisted driving technologies puts forward some new challenges to the safety of intersection. To improve the prediction of collision risk at intersections, this study proposes a collision prediction method that integrates the driving behavior entropy with a bidirectional Long Short-Term Memory (Bi-LSTM) network. Based on a simulation platform of high-fidelity driving, three typical conflict scenarios at urban intersections were constructed: the conflicts of right-turn pedestrians and left-turn oncoming vehicles, and straight-driving obstacle avoidance. The data of longitudinal speed and lateral steering wheel angle control behavior were collected from 55 drivers. A sliding time window, speed entropy, steering wheel angle entropy, and their corresponding standard deviations were calculated to construct the feature indicators of multidimensional driving stability. Repeated-measures analysis of variance (ANOVA) was employed to statistically analyze the feature differences across different scenarios and driving phases. A Bi-LSTM model was further developed to predict the collision risk in real time within a 3-second time window. The results indicate that both standard deviation-based indicators and driving behavior entropy measures during conflict phases were significantly higher than those during the stable driving phases, and entropy indicators effectively captured the uncertainty and complexity of driver operations. The incorporation of entropy features reduced the miss rate by 30.07% and the false alarm rate by 42.45%, while it improves the average early warning time by 36.84%. As an effective complement to traditional fluctuation- based indicators, the entropy of driving behavior can enhance the sensitivity and prediction stability of model under hazardous conditions. It provides a feasible reference method for intersection risk warning in the assisted driving environments and contributes to the improvement of road traffic safety.

Key words: intelligent transportation, risk prediction, driving behavior entropy, assisted driving