2I-PBL: A Hybrid Stochastic Car-Following Model Integrating PSO-Bi-LSTM and Stochastic 2D-IDM
IEEE Access · 2026 · 14 · 111930-111943
DOI: 10.1109/access.2026.3713687 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Zhenming Yang und 1 weiteren Autor:innen. Die Arbeit nennt 57 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Opening Foundation of Jiangsu Province Engineering Research Center of Smart Poultry Farming and Intelligent Equipment.
Abstract (Originalsprache)
Traffic congestion frequently emerges in daily transportation systems due to complex and heterogeneous car-following (CF) behaviors, making accurate and reliable CF modeling a longstanding challenge in micro traffic flow research. To address the trade-off between predictive accuracy, interpretability, and safety in existing CF models, this study proposes a hybrid stochastic car-following framework termed the 2I-PBL model. The proposed model integrates a data-driven PSO-Bi-LSTM model with a theory-driven stochastic 2D Intelligent Driver Model (2D-IDM). Specifically, particle swarm optimization (PSO) is first employed to optimize the hyperparameters of the Bi-directional Long Short-Term Memory (Bi-LSTM) network, yielding a PSO-Bi-LSTM model with enhanced prediction accuracy. Subsequently, a combined forecasting strategy is introduced to fuse the optimized PSO-Bi-LSTM model with the stochastic 2D-IDM, enabling the hybrid model to incorporate both data-adaptive learning capability and physically interpretable driving mechanisms while accounting for stochastic human driving behaviors. Simulation experiments demonstrate that the proposed 2I-PBL model achieves superior predictive performance compared with conventional deterministic and single-model approaches, while maintaining safety constraints and physical consistency. Moreover, the model effectively captures heterogeneous driving behaviors under varying traffic conditions. The results indicate that the proposed framework provides a promising solution for bridging data-driven and theory-driven CF modeling, offering improved accuracy, interpretability, and robustness, and contributing to the development of safer and more realistic microscopic traffic flow models.
Förderung
- Opening Foundation of Jiangsu Province Engineering Research Center of Smart Poultry Farming and Intelligent Equipment
Zitieren
APA
Yang, Z., & Guo, S. (2026). 2I-PBL: A Hybrid Stochastic Car-Following Model Integrating PSO-Bi-LSTM and Stochastic 2D-IDM. IEEE Access, 14, 111930-111943. https://doi.org/10.1109/access.2026.3713687
Vancouver
Yang Z, Guo S. 2I-PBL: A Hybrid Stochastic Car-Following Model Integrating PSO-Bi-LSTM and Stochastic 2D-IDM. IEEE Access. 2026;14:111930-111943. doi:10.1109/access.2026.3713687
BibTeX
@article{Yang2026,
author = {Yang, Zhenming and Guo, Shengzhuo},
title = {2I-PBL: A Hybrid Stochastic Car-Following Model Integrating PSO-Bi-LSTM and Stochastic 2D-IDM},
journal = {IEEE Access},
year = {2026},
volume = {14},
pages = {111930--111943},
doi = {10.1109/access.2026.3713687}
}
EUVIMED