Learning Based Security for VANET with Blockchain

The security issue is one of the greatest challenges in vehicular ad hoc networks (VANETs) attracting a great deal of attention. Malicious onboard units (OBUs) can attack other OBUs with various manners to obtain illegal gains, such as jamming, eavesdropping spoofing and so on. To reduce the potential attackers in the network, we propose an indirect reciprocity security framework with a scalar reputation assigned to each OBU to evaluate their dangerous level to the VANET. A blockchain technique that uses consensus mechanism and encryption algorithms to protect information from being tampered is applied for the transmitter to record the behaviors of other OBUs. We also propose a reinforcement learning based action selection strategy for an OBU in the VANET to choose a reliable relay OBU or determine whether to follow the request of a source OBU or not. A hotbooting technique is applied for the OBUs with prior knowledge to accelerate the learning speed. Simulation results show that the proposed action selection strategy can efficiently increase the packet delivery ratio, the reputation and the utility of the each OBU.