Quantifying and predicting driving risks for connected autonomous vehicles (CAVs) is critical to ensuring the safe operation of traffic in complex environments. This study first establishes a car-following model for CAVs based on molecular force fields. Subsequently, using a convolutional neural network and long short-term Memory (CNN-LSTM) deep-learning model, the future trajectory of the target vehicle is predicted. Risk is quantified by employing models that assess both the collision probability and collision severity, with deep-learning techniques applied for risk classification. Finally, the High-D dataset is used to predict the vehicle trajectory, from which the speed and acceleration of a target vehicle are derived to forecast driving risks. The results indicate that the CNN-LSTM model, when compared with standalone CNN and LSTM models, demonstrates a superior generalization performance, a higher sensitivity to risk changes, and an accuracy rate exceeding 86% for medium- and high-risk predictions. This improved accuracy and efficacy contribute to enhancing the overall safety of connected vehicle platoons.

Connected autonomous vehicles (CAVs) face constraints from multiple traffic elements, such as the vehicle, road, and environmental factors. Accurately quantifying the vehicle’s operational status and driving risk level in complex traffic scenarios is crucial for enhancing the efficiency and safety of connected autonomous driving. To continuously and dynamically quantify the driving risks faced by CAVs in the road environment—arising from the front, rear, and lateral directions—this study focused s on the self-driving particle characteristics that enable CAVs to perceive their surrounding environment and make driving decisions. The vehicle-to-vehicle interaction behavior was analogized to the inter-molecular interaction relationship, and a molecular Morse potential model was applied, coupled with the vehicle dynamics theory. This approach considers the safety margin and the specificity of driving styles. A multi-layer decoder–encoder long short-term memory (LSTM) network was employed to predict vehicle trajectories and establish a risk quantification model for vehicle-to-vehicle interaction behavior. Using SUMO software (win64-1.11.0), three typical driving behavior scenarios—car-following, lane-changing, and yielding—were modeled. A comparative analysis was conducted between the risk field quantification method and existing risk quantification indicators such as post-encroachment time (PET), deceleration rate to avoid crash (DRAC), modified time to collision (MTTC), and safety potential fields (SPFs). The evaluation results demonstrate that the risk field quantification method has the advantage of continuously quantifying risk, addressing the limitations of traditional risk indicators, which may yield discontinuous results when conflict points disappear. Furthermore, when the half-life parameter is reasonably set, the method exhibits more stable evaluation performance. This research provides a theoretical basis for the dynamic equilibrium control of driving risks in connected autonomous vehicle fleets within mixed-traffic environments, offering insights and references for collision avoidance design.