ARSL-V: A Risk-aware Relay Selection Scheme using Reinforcement Learning in VANETs

In high-speed and dynamic Vehicular Ad-hoc Networks (VANETs), cooperative transmission mechanism is a promising scheme to ensure the sustainable transmission of data. However, due to the possible malicious behavior of vehicles and the dynamic network topology of VANETs, not all vehicles are trustworthy to become relays and perform the cooperative transmission task reliably. Therefore, how to ensure the security and reliability of the selected vehicles is still an urgent problem to be solved in the relay selection research area. In this paper, we propose a risk-aware relay selection scheme (ARSL-V) using reinforcement learning in VANETs. Specifically, we design a risk assessment mechanism based on multiple parameters to dynamically assess the potential risk of relay vehicles by considering the reputation variability, abnormal behavior, and environmental impact of vehicles. Also, we model the relay selection problem as an improved Kuhn-Munkres algorithm based on the risk assessment to realize relay selection in multi-relay and multi-target vehicle scenarios. Besides, we use a reinforcement learning algorithm combined with feedback data to achieve dynamic adjustment of parameter weights. Simulation results show that compared with the existing schemes, ARSL-V can improve the detection rate of malicious behavior and cooperative transmission success rate by about 25% and 6%, respectively.