LEARNING AND MANAGING STOCHASTIC NETWORK TRAFFIC DYNAMICS: A DIGITAL TWIN BASED APPROACH

This study generates understanding on the feasibility of adopting the digital twin-based approach to optimize congestion tolling strategies with the consideration of time-varying traffic dynamics. Particularly, a truth model is developed to mimic the traffic environment and utilize it as the physical entity for digital twinning. A twin model that mirrors the truth model is formulated and calibrated for optimizing and training congestion tolling adjustment strategies with the help of Reinforcement learning techniques. The optimized prices are then input into the "truth model" to test the performance of enhancing the network efficiency. The above procedure is iterative, which can be applied for the congestion management with the period-toperiod tolling adjustment scheme in practice. Extensive experiments are conducted to illustrate that the proposed digital twin-based tolling adjustment scheme is able to enhance the overall efficiency of transport networks, which enlightens a novel direction on congestion management by digital twin-based approaches.