A Dynamic Pricing Model for an Online-selling Parking Permits - Reinforcement Learning Approach

This study considers a situation where a parking lot sells a 3-hour parking permit online with a static price through an O2O platform. Instead of the static pricing policy, the parking lot aims to dynamically control the parking permit price to prevent it from being full and getting too overcrowded. To address this problem, we propose a DQN(Deep Q-Network)-based dynamic pricing model that determines the price for every hour after observing the parking environment such as occupancy and the amount of sold permits. We trained the DQN model with an environment that simulates demands, incoming and outgoing vehicles, parking duration and lead time. Numerical analysis compares the proposed dynamic pricing model with the current static pricing policy. The comparison demonstrates that the dynamic pricing model is effective for reducing the chance of overcrowded and enhancing the profitability.