An Optimal Packet Delivery Strategy Based on Deep Reinforcement Learning in IoV

Abstract

Internet of Vehicles (IoV) have been widely regarded as a promising technology to implement the intelligent transportation system. The implementation of IoV is inseparable from Vehicle-to-Vehicle (V2V) communication and Vehicle-to-Everything (V2X) communication. There have been many studies on packet delivery strategy in IoV, but most of all these studies ignore the links between the characteristics of the vehicle routing, road topology and communication. Therefore the performance of IoV communication is still not guaranteed. This paper proposes a packet delivery strategy based on deep reinforcement learning in IoV, which considers vehicles position, topology, the successful delivery ratio, congestion probability and the life cycle of packets. Firstly, we establish and optimize the packet delivery model. Then, based on the deep reinforcement learning method, we propose our packet delivery strategy. We also design and realize the packet delivery system in IoV with Simulation of Urban Mobility (SUMO) to obtain the real movement model of vehicles. Furthermore, we analyze the performance of our strategy compared to existing methods in packet delivery. The experimental results demonstrate that the proposed model improves the efficiency of packet delivery transmission and reduce the congestion under different network communication loads.