HTSA-LSTM: Leveraging Driving Habits for Enhanced Long-Term Urban Traffic Trajectory Prediction

• DOI: 10.20944/preprints202501.0717.v2
• Publication Date: 2025-01-14T03:20:53Z
• Authors: Yiying Wei; Xiangyu Zeng; Xirui Chen; Hui Zhang; Hui Wang; Zhengan Yang; Bingming Jia; Zicheng Li
• Keywords: Adaptability; Benchmark (surveying); Robustness

The rapid evolution of intelligent vehicle technology has significantly advanced autonomous decision-making and driving safety. However, the challenge predicting long-term trajectories in complex urban traffic persists, as traditional methodologies usually handle spatiotemporal attention mechanisms isolation are typically limited to short-term trajectory predictions. This paper proposes a Habit-based Temporal-Spatial Attention Long Short-Term Memory (HTSA-LSTM) network, novel framework that integrates dual mechanism capture dynamic dependencies across time space, coupled with style analysis module. module employs Sparse Inverse Covariance Clustering Spectral (SICC-SC) extract primitives cluster data, thereby revealing diverse behavior patterns without relying on predefined labels. By segmenting real-world data into fundamental behavioral units reflect individual preferences, this approach enhances model's adaptability. These units, conjunction outputs, serve inputs model, ultimately improving prediction accuracy robustness multi-vehicle scenarios. model was evaluated by using NGSIM dataset real from Wuhan, China. In comparison benchmark models, HTSA-LSTM achieved 20.72% reduction root mean square error (RMSE) 24.98% negative log likelihood (NLL) for 5-second predictions trajectories. Furthermore, R² values exceeding 97.9% highways expressways over 92.7% 3-second roads, highlighting its excellent performance adaptability conditions.