Dynamic Vehicle Data Gathering via Deep Reinforcement Learning Approach

Abstract

With the rapid development of vehicular ad hoc networks (VANETs), there have been numerous efforts to big data and analysis of vehicle information for roadside intelligence. However, continuous data gathering is energy consuming and eventually causes data backlog due to the capacity-limited backhaul links, while sparse gathering frequency may miss the timely detection of critical traffic information. Therefore, this paper focus on the dynamic data gathering problem in vehicular networks. In the scenario with environment changes dynamically, we first model the problem as a Markov decision process (MDP), and then propose different deep reinforcement learning(DRL) based maximization frequency matching algorithms, to determine the optimal gathering frequency at each time. The simulation results compare the performance differences of the algorithms, and show the trend of frequency matching in different storage spaces.