Sun Guanghui, Zhang Hongbin, Zhong Liande, Li Qingqing
School of Transportation and Vehicle Engineering, Shandong University of Technology, Zibo, 255000, China.
Research Institute of Highway Ministry of Transport, Beijing, 100089, China.
Sci Rep. 2025 Mar 27;15(1):10631. doi: 10.1038/s41598-025-94000-6.
The driving environment in continuous diverging sections of expressway system interchanges is highly complex, posing significant driving risks. To investigate driving behavior and its transition patterns in these areas, a simulated driving experiment was conducted to collect driving behavior parameters and construct a driving behavior spectrum (DBS) for continuous diverging sections. A driving behavior spectrum unit decomposition model, leveraging convolutional neural networks (CNN) and bidirectional long short-term memory networks (BiLSTM), was developed to identify specific driving behaviors. Additionally, a Hidden Markov Model (HMM) was employed to quantify transitions between various driving behavior states. The findings demonstrate that the DBS effectively captures and systematically records temporal changes in driving behavior. The CNN-BiLSTM model accurately identified four typical driving behaviors-straight driving, lane changing, deceleration, and turning-with an impressive average accuracy of 98%. Analysis revealed that the first lane change typically occurs approximately 121 m before the first diverging point, while the second occurs around 78 m before the second diverging point. Furthermore, the HMM model successfully elucidated the transition patterns between different driving states. These results provide valuable insights for identifying hazardous zones and optimizing facility design in expressway system interchanges.
高速公路系统互通式立体交叉连续分流路段的驾驶环境极为复杂，存在重大驾驶风险。为了研究这些区域的驾驶行为及其转变模式，开展了一项模拟驾驶实验，以收集驾驶行为参数并构建连续分流路段的驾驶行为谱（DBS）。开发了一种利用卷积神经网络（CNN）和双向长短期记忆网络（BiLSTM）的驾驶行为谱单元分解模型，以识别特定的驾驶行为。此外，采用隐马尔可夫模型（HMM）来量化各种驾驶行为状态之间的转变。研究结果表明，驾驶行为谱有效地捕捉并系统记录了驾驶行为的时间变化。CNN-BiLSTM模型准确识别出四种典型驾驶行为——直线行驶、变道、减速和转弯，平均准确率高达98%，令人印象深刻。分析表明，第一次变道通常发生在第一个分流点前约121米处，而第二次变道则发生在第二个分流点前约78米处。此外，隐马尔可夫模型成功阐明了不同驾驶状态之间的转变模式。这些结果为识别危险区域和优化高速公路系统互通式立体交叉的设施设计提供了有价值的见解。
