Dynamic assignment problem of parking slots

The parking problem has become one of the major issues in urban transportation planning and traffic management research. The present paper deals with the dynamic assignment problem of the parking slots (places). The objectives are to provide a global satisfaction of all customers and maximize parking occupancy. A dynamic assignment problem consists in solving a sequence of assignment problems over time. To cope with all these aspects, we offer in this paper, a new approach based on a learning strategy: an Estimation of Distribution Algorithm (EDA) where a reinforcement learning method is used to support the assignment algorithm. We tested our approach with simulations for over 120 days, using a set of up to 10 parking lots (i.e. with up to 7,000 parking slots) and 13,000 requests distributed with different patterns over a day. The comparative study between an assignment algorithm with and without reinforcement learning algorithm has proven the relevance of our approach: the saving is up to 80%. The results also showed the effects and the benefits of the learning strategy.