Prof. Dr. Kenan Zhang
EPFL
Title: Markov population potential games: An alternative framework for Markovian traffic assignment
Traffic assignment models describe how selfish individuals interact on a physical network under prescribed behavioral rules, and how an equilibrium state emerges from these interactions. Although originally developed for long-term travel demand forecasting, their applications extend far beyond road traffic prediction. Among existing models, Markovian traffic assignment (MTA) strikes a balance between model representational power and analytical tractability by assuming memoryless (Markovian) system dynamics and state-dependent decision making. In this talk, I will introduce a new framework for Markovian traffic assignment, namely, Markov population potential game (MPPG), which integrates Markov game, potential game, and population game. By allowing introducing interdependent state transitions, MPPG generalizes classical MTA and substantially broadens the scope of applicability. A key advantage of MPPG, as with potential games in general, is the existence of an equivalent optimization problem whose optimal solution corresponds to the equilibrium. I will present sufficient conditions under which such a potential exist and can be constructed in closed form, along with solution algorithms based on the corresponding optimization formulation, then conclude with several applications of MPPG in transportation.