Mean-Field Learning for Day-to-Day Departure Time Choice with Mode Switching

Ben Wang, Qi Luo, Yafeng Yin

Abstract: Understanding travelers' day-to-day departure time choice (DDTC) is vital for managing traffic congestion, especially in multi-modal transportation systems. While providing real-time traffic information and alternative trip plans brings convenience to travelers, their collective travel patterns may conversely lead to unstable traffic equilibrium states. We investigate a DDTC problem with mode switching in this paper. A group of heterogeneous agents can adaptively choose their modes and departure times to minimize total travel costs in a dynamic game. Using a customized hierarchical soft actor-critic (HSAC) algorithm with a continuum approximation of other agents, the traffic dynamics will converge to an approximate Markovian Perfect Equilibrium (MPE). Our findings also shed light on changes in long-term travel behavior due to the widespread deployment of emerging mobility and travel information technology. This approach serves as a foundation for promoting intelligent travel plans through adaptive traffic control policies.