This research focuses on the understanding of driving behaviour in motorway weaving sections, particularly the lane-changing and acceleration behaviours which are significant factors in characterising the operations of weaving section.
Drivers’ lane-changing behaviour is a series interdependent decisions according to a particular lane-changing plan (latent). An intensive interaction with neighbouring traffic increases the lane-changing complexity in weaving section. The drivers’ choices in weaving section can be significantly affected by the actions of the neighbourhood drivers and moving as a group (i.e. platoon and weaving). Furthermore, the intensity of lane-changing has significant impact on the acceleration behaviour in weaving section traffic which may response differently from the stimulus (i.e. leave a space for pre-emptive lane-changing).
An analysis of detailed trajectory data collected from moderately congested traffic flow of a typical weaving section in the M1 motorway, UK (J 42-43). The data reveals that a substantial proportion (23.4%) of the lane-changing at weaving section exhibits such group behaviour (i.e. platoon and weaving). The current study extends the state-of-the-art latent plan lane-changing model which account explicitly the various mechanisms. The model constitutes that the driver is most likely performing a pre-emptive lane-changing at the beginning of weaving section and moving toward kerbside (left direction). Moreover, the driver aggresiveness affects significantly on weaving and least on platoon lane-changing.
The proposed acceleration model allows the car-following behaviour (acceleration/deceleration) corresponds with both stimulus (positive/negative relative speed). The model is conditional on gap threshold and reaction time distributions (probabilistic model) capturing the heterogeneity across drivers. most of traffic response differently from the stimulus condtions where 43.5% falls in deceleration with positive relative speed.
All the parameters in each model are estimated jointly using Maximum Likelihood Estimation technique and reveal significant differences. The results show promising contribution towards improving the fidelity of microscopic traffic performance analysis.
