Impact of Lane-Change Decision Variations of Autonomous Vehicles on Overall Traffic Efficiency in the Autonomous Traffic Environment

To investigate the impact of variations in lane-change decision-making strategies of autonomous vehicles on the overall efficiency of merging areas, decision models incorporating different driving styles are developed for further analysis. First, by analyzing the differences in mandatory lane-change behavior, model parameters for various driving styles are designed. A lane-change decision model based on Double Deep Q-Network (DDQN) is then developed using reinforcement learning (RL). This model integrates three key aspects: efficiency, safety, and merging tasks, in the design of a reward function aimed at optimizing the lane-change decision-making process. A simulation platform for the merging area scenario is constructed using SUMO (Simulation of Urban Mobility) to evaluate the performance of the lane-change decision model. The experimental results demonstrate that the model performs effectively across various metrics, including task success rate and vehicle speed. Furthermore, through an analysis of the impact of variations in autonomous vehicle lane-changing decisions on traffic efficiency under different traffic volumes, this study finds that congestion in merging areas progressively worsens as traffic flow increases. Additionally, from a holistic perspective, vehicles employing the normal strategy are less likely to induce traffic congestion during the merging process compared to those utilizing aggressive or conservative strategies, and their speed distribution tends to be higher than that of the other two strategies.