Reinforcement Learning based Gateway Selection in VANETs

Abstract

In vehicular ad hoc networks (VANETs), providing the Internet has become an urgent necessity, where mobile gateways are used to ensure network connection to all customer vehicles in the network. However, the highly dynamic topology and bandwidth limitations of the network represent a significant issue in the gateway selection process. Two objectives are defined to overcome these challenges. The first objective aims to maximize the number of vehicles connected to the Internet by finding a suitable gateway for them depending on the connection lifetime. The second objective seeks to minimize the number of connected vehicles to the same gateway to overcome the limitation of gateways' bandwidth and distribute the load in the network. For this purpose, A gateway discovery system assisted by the vehicular cloud is implemented to find a fair trade-off between the two conflicting objectives. Proximal Policy Optimization, a well-known reinforcement learning strategy, is used to define and train the agent. The trained agent was evaluated and compared with other multi-objective optimization methods under different conditions. The obtained results show that the proposed algorithm has better performance in terms of the number of connected vehicles, load distribution over the mobile gateways, link connectivity duration, and execution time.