交叉口直行车辆出口车道选择行为的贝叶斯网络模型构建

To reveal the influence of intersection lane layout, traffic flow rate, and vehicle location on the behavior of exit lane selection, to safeguard the operation of straight-through traffic, this paper builds a Bayesian Network (BN) model for straight-through exit lane selection. Firstly, given the insufficient description of the vehicle location in the existing straight-through exit-lane selection model, the paper adds the influencing factors such as whether the vehicle arrives during the green signal countdown display and whether it is the first vehicle. They characterize the spatiotemporal locations of different vehicles within the same cycle. Then, knowledge and correlation analysis are applied to construct a network structure for 10 influencing factors such as the centerline offset of the innermost straight lane, and the number of lanes. Data-driven method is used to complete parameter learning. The Mean Absolute Error (MAE) and Root Mean Square Deviation (RMSE) of the model are verified to be 4. 37% and 4. 96%, which have good prediction accuracy. Applying forward reasoning gives (1) the position of the entrance lane, the matching of the number of lanes, the mismatching ratio of the choice of the entrance and exit lanes, and the centerline offset of the innermost straight lane are the main factors affecting the selection behavior of the straight exit-lanes. (2) When the centerline offset of the innermost straight lane is below 4° and the average flow rate of the entrance lane is 600 900 veh/ (h·lane), the mismatching ratio tends to stay between 25% and 50% . (3) When the centerline offset is below 6° and the number of lanes is matched, the mismatching ratio tends to stay between 50% and 75% . (4) When the centerline offset is above 6° and the number of lanes is not matched, the mismatching ratio tends to stay above 75% . The paper exerts scenario analysis method to predict that, e. g. (5) When the average straight traffic flow is 300-600 veh/ (h·lane), the headway maintains a higher speed through an intersection with the mismatched number of lanes and an offset of 6° or more, and the utilization of the innermost exit lane is only 10. 4% . (6) Under the conditions that the intersection straight-through traffic flow is 600-900 veh/ (h · lane), Green Signal Countdown Display (GSCD) shows arrival, and the centerline offset of the innermost straight lane is 4° or less, the probability that the straight-through vehicle in the innermost entrance lane chooses the innermost exit lane is 44. 6% . It can provide traffic managers with prediction tools for direct exit lane selection and help with theoretical studies on the safety of straight-through traffic flow.