Spatiotemporal Characteristics of Longitudinal Risky Driving Behaviors in Highway Tunnel Sections

To pinpoint the patterns, locations, and timings of longitudinal risky driving behaviors in tunnel sections, enhancing the ability of traffic management departments to proactively prevent accidents, this study addresses the limitations of conventional separate spatio-temporal analysis dimensions by developing a spatio-temporal kernel density estimation model (STKDE). The model's optimal bandwidth is determined using least squares cross-validation (LSCV).A method for identifying longitudinal risky driving behaviors based on trajectory data is constructed, extracting spatio-temporal locations for four risky driving behaviors: speeding, extreme low speed, rapid acceleration, and rapid deceleration. By partitioning the spatio-temporal domain of the study area into units, STKDE is applied to compute the spatio-temporal kernel density estimation value, ψ, within each unit. The results of STKDE are visualized using a space-time cube model (ST-Cube). An empirical analysis based on high-precision trajectory data from the Xiaxiyao Tunnel reveals that high-speed driving behavior frequently occurs within 100 meters of the tunnel exit, with speeding peaking at 16:00 and 09:00. Low-speed driving behavior is frequent within 200 meters before the tunnel entrance, with extreme low speed peaking at 02:00 and 14:00. Within 100 meters before entry and throughout the first 1500 meters of the tunnel, the ψ values for rapid acceleration and deceleration remain above 0.5, indicating high-frequency occurrences.. Additionally, every 150~200 meters within the tunnel, these two types of sudden speed changes show simultaneous fluctuations, but significantly decrease and are no longer frequent once exiting the tunnel. A comparison with conventional spatio-temporal analysis methods shows that the STKDE method, combined with ST-Cube, achieves integrated spatio-temporal feature analysis and provides a quantifiable estimation of the likelihood of risky driving behaviors across the entire spatio-temporal domain, demonstrating a particular advantage in characterizing rapid acceleration and deceleration behaviors.