Mobile parking incentives for vehicular networks: a deep reinforcement learning approach

In vehicular networks, parked vehicles can join vehicular communication as static nodes, but encouraging people to share their wireless devices during parking still suffer from user selfishness, unexpected parking locations, and delayed data transmission. In this study, we propose a centralized service provider (SP) system to distribute mobile parking incentives for vehicle users, as some parking tasks via a smartphone app. With the support of SP system, mobile users can accept parking tasks and park their vehicles on proposed locations for rewards, which are beneficial to vehicular communication. Meanwhile, a dynamic pricing algorithm based on reinforcement learning (RL) for SP system that aiming to maximize the connectivity of all road segments and minimize the total cost of the SP is considered. Since SP system has a slow learning speed in the presence of a large state space due to the curse of high dimensionality, we develop a deep Q-network (DQN) based payment strategy that exploits a deep neural network to compress the learning state space and estimate the Q-value for each payment value. Extensive simulations with realistic parking deployments demonstrate that the proposed scheme accelerates the learning speed to obtain an optimal payment policy and increases the completion rate of tasks, and the final parking results effectively enhance vehicular communication.