Long-Term Trajectory Prediction Method Based on Highway Vehicle-Following Behavior Patterns

Abstract

To address existing shortcomings such as short time domains and low interpretability, this study proposes a long-term trajectory prediction model for leading vehicles that considers the impact of traffic flow. Through an analysis of trailing trajectory data from the HighD natural driving dataset, fitting relationships for the following behavior patterns were derived. Building upon the intelligent driver model (IDM), three long-term trajectory prediction models were established: acceleration delta velocity (ADV), space delta velocity intelligent driver model (SDVIDM), and space velocity intelligent driver model (SVIDM). These models were then compared with the IDM model through simulations. The results indicate that when there is one vehicle ahead, under aggressive following conditions, the ADV model outperforms the IDM model, reducing the root mean square errors in acceleration, speed, and position by 79.61%, 91.26%, and 87.82%, respectively. In scenarios with two vehicles ahead and conservative short-distance following, the SDVIDM model exhibits reductions of 83.42%, 92.85%, and 92.25%, while the SVIDM model shows reductions of 82.31%, 92.47%, and 94.02%, respectively, compared to the IDM model.