Physics-Informed Deep Learning for Traffic State Estimation with Varying Speed Limits
A new framework has been introduced that combines teacher-student ensemble training with physics-informed deep learning (PIDL) neural networks for traffic state estimation (TSE) in scenarios with varying speed limits (VSL). Current PIDL models struggle to adapt to fluctuating traffic conditions on freeways with VSL. The innovative method incorporates the principles of flow conservation law in teacher models through PIDL, while a student model employs a multi-layer perceptron classifier (MLP) to recognize traffic patterns and choose the suitable ensemble member. This advancement tackles a significant shortcoming in existing traffic management strategies.
Key facts
- Physics-informed deep learning (PIDL) neural networks are used for traffic state estimation (TSE).
- Varying speed limits (VSLs) are an efficient traffic management approach.
- Existing PIDL training architectures cannot handle changing traffic characteristics on freeways with VSL.
- A novel framework integrates teacher-student ensemble training with PIDL for TSE under VSL scenarios.
- Teacher models encode the physics of flow conservation law locally using PIDL.
- The student model uses a multi-layer perceptron classifier (MLP) to identify traffic characteristics.
- The student model selects the ensemble member of PIDL neural networks.
- The framework is proposed to tackle the challenge of VSL in traffic state estimation.
Entities
Institutions
- arXiv