Expressway dynamic speed limit control based on physics-informed neural network

ObjectiveTraditional data-driven models suffer from insufficient physical consistency, which are difficult to adapt to complex traffic scenarios. This study proposes a dynamic speed limit control method based on physics-informed neural network (PINN) to achieve the synergistic optimization on data fitting and physical laws, thereby providing reliable support for intelligent traffic management and control.MethodA coupled modeling framework integrating data and physical laws was constructed by embedding Lighthill-Whitham-Richards macroscopic traffic flow equation, as a physical constraint, into the loss function of neural network. The sensitivity parameter optimization was employed to balance the weights of data fitting and physical constraints. An automatic differentiation mechanism was combined to dynamically calculate derivatives, improving the generalization ability of model and the accuracy of depicting traffic flow evolution.ResultThe verification based on field-measured data from Daqing-Guangzhou expressway and UC-win/Road simulation shows that compared with the fixed speed limit strategy, the proposed method reduces congestion time by 38.73%, increases traffic volume by 21.43%, and decreases speed variance by 55.44%. It outperforms mainstream data-driven models. It reduces congestion time by 26.27%, increases traffic volume by 10.12%, and decreases speed variance by 33.85% compared with LSTM-MPC model. It reduces congestion time by 13.86%, increases traffic volume by 5.06%, and decreases speed variance by 28.65% compared with deep reinforcement learning model.ConclusionThe proposed method effectively integrates data characteristics with traffic flow physical laws, exhibiting excellent physical interpretability and control effectiveness. It provides a new data-model collaborative driven approach for intelligent traffic dynamic control and is of great significance for improving intelligent management and control level of expressways.