Conditional probability of node .

To dynamically identify road section risks, the research team designed an onboard unit to collect various dynamic driving behavior data when the Advanced Driver Assistance Systems (ADAS) are activated. To do this, we divided the roads into three categories (urban road, expressway, freeway) and established separate BN models to analyze the relationship between driving behavior and road section risk. These models were constructed based on natural driving data from 10,000 km, collected from vehicles equipped with ADAS. For road segment division, fixed-length intervals were used for freeways and urban expressways, while segments on urban roads were defined as the stretches between adjacent intersections. Using braking deceleration and time to collision, we identified near-crash events and classified them into high, medium, and low severity levels using the DBSCAN clustering algorithm. These near-crash events were then matched to the corresponding road section, assigning different weights based on their severity levels to evaluate the risk level of each segment. Additionally, driving behavior data, including velocity, lateral acceleration, longitudinal acceleration, yaw rate, accelerator position, steering angle, and steering angle velocity, were matched to the road segments. Finally, using the Netica software, Bayesian network models were constructed separately for urban roads, expressways, and freeways to identify driving risks at road segments. The models exhibited high sensitivity to observed nodes.

为动态识别道路路段风险，本研究团队设计了车载单元，用于在高级驾驶辅助系统（Advanced Driver Assistance Systems，ADAS）激活时采集各类动态驾驶行为数据。为此，我们将道路划分为三类：城市道路、快速路与高速公路，并构建了独立的贝叶斯网络（Bayesian Network，BN）模型，以分析驾驶行为与路段风险之间的关联。上述模型依托总里程达10000公里的自然驾驶数据构建，该数据采集自配备ADAS的车辆。在路段划分方面，高速公路与城市快速路采用固定长度间隔进行分段，而城市道路的路段则定义为相邻交叉口之间的路段。研究人员利用制动减速度和碰撞时间识别险态事件，并通过基于密度的带噪声应用空间聚类（Density-Based Spatial Clustering of Applications with Noise，DBSCAN）算法将其划分为高、中、低三个严重程度等级。随后将这些险态事件匹配至对应的路段，并依据其严重程度赋予不同权重，以评估各路段的风险等级。此外，研究人员还将速度、横向加速度、纵向加速度、横摆角速度、油门开度、转向角及转向角速度等驾驶行为数据匹配至各路段。最终，借助Netica软件，分别针对城市道路、快速路及高速公路构建贝叶斯网络模型，以识别路段驾驶风险。该模型对观测节点表现出较高的敏感性。