Dynamic Multiobjective Optimization of Lane‐Changing Trajectories Based on Reinforcement Learning

To enhance the rationality of lane‐changing decisions and the adaptability of trajectory planning, this study incorporates short‐term driving styles to construct a multiobjective optimized lane‐changing trajectory planning model based on naturalistic driving data. First, lane‐changing behavior rules were defined to extract lane‐changing and lane‐keeping data. Essential factors influencing lane‐changing behavior were identified using the eXtreme Gradient Boosting (XGBoost) model. Based on the essential factors, drivers were classified into three categories (conservative, moderate, and aggressive) using Density‐Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm, which revealed the behavioral difference during lane‐changing. Subsequently, an attention‐enhanced long short‐term memory (LSTM) network was employed to predict surrounding vehicle trajectories, generating dynamically updated environmental parameters. Further considering comfort and safety benefits during lane‐changing, a multiobjective trajectory planning model was developed. Reinforcement learning algorithms iteratively optimized the trajectories to derive the optimal trajectory. Finally, the behavioral characteristics of planned trajectories for the three categories of drivers and the deviations between planned and actual trajectories were compared. Results indicate that planned trajectories exhibit shorter lane‐changing length and higher efficiency compared with actual trajectories. Planned trajectory can smooth microlevel behavior and improve safety and comfort during lane‐changing. For different types of drivers, conservative drivers show the longest lane‐changing length but smallest headway space distances, which reflects drivers’ caution during lane‐changing. Aggressive drivers mostly focus on speed improvement. The findings can be applied to vehicle trajectory planning in connected environments, which can enhance the lane‐changing efficiency while ensuing safety.