A method and system for controlling ramp traffic on a highway based on physical information reinforcement learning
By combining the METANET physical model and the TD3 algorithm, and introducing a physical loss term and a hybrid experience replay mechanism, the problems of model error and reinforcement learning instability in highway ramp control are solved, realizing collaborative optimization control of ramp traffic flow and improving the operational efficiency and stability of the traffic system.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-03-27
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies for highway ramp control suffer from model errors and parameter mismatches, making it difficult to balance real-time performance and robustness in complex scenarios. Traditional reinforcement learning, on the other hand, faces problems such as low sample efficiency, unstable training, and insufficient policy interpretability.
By combining the METANET macroscopic traffic flow physical model with the TD3 algorithm, and by introducing a physical loss term and a hybrid experience replay mechanism, the reinforcement learning model is trained by integrating real and virtual experiences, ensuring the consistency and interpretability of the strategy.
It improves sample efficiency and training stability, generates stable and efficient ramp control strategies in complex traffic scenarios, reduces total time delay, and improves the operational efficiency and stability of the traffic system.
Smart Images
Figure CN122416756A_ABST