Optimizing longitudinal control model parameters of connected and automated vehicles using empirical trajectory data of human drivers in risky car-following scenarios

Connected and automated vehicles (CAVs) have great potential to improve driving safety. A basic performance evaluation criterion of CAVs is whether they can drive more safely than human drivers in real traffic scenarios. This study proposes a method to optimize longitudinal control model parameters of CAVs using empirical trajectory data of human drivers in risky car-following scenarios. Firstly, the initial car-following pairs (I-CFP) are extracted from empirical trajectory data. Then, two types of real longitudinal control models of CAVs, the adaptive cruise control (ACC) and the cooperative ACC (CACC) control models, are employed for simulation in the car-following scenarios with default parameter values, which generate original trajectories of simulated car-following pairs (S-CFP). Finally, a genetic algorithm (GA) is applied to optimize control model parameters of ACC and CACC vehicles and generate optimized trajectories of car-following pairs (O-CFP). Results indicate that safety condition of S-CFP is better than that of I-CFP, while the O-CFP has the best safety performance. The optimized parameters in the ACC/CACC models are diverse and different from the default parameters, indicating that the best model parameters vary with different car-following scenarios. Findings of this study provide a valuable perspective to reduce the rear-end collision risks.