| With the rapid increase of vehicle ownership,traffic congestion has brought huge traffic congestion to cities.Thanks to the rapid development of intelligent technology and wireless communication technology,Intelligent Transportation System(ITS)is considered to be an effective way to alleviate traffic congestion.However,there are huge network security risks in the extensive connection between vehicles and infrastructure.Malicious signal attackers can send forged Connected Vehicle(CV)trajectory information to Road Side Unit(RSU)and signal controller to mislead the controller’s perception and decision-making,which may wipe out the advantages of ITS.How to evaluate and defend against the signal attack that the attacker may launch has become an urgent problem to be solved.In the past,the research of signal attack only focuses on the identified controller or the known algorithm source code.However,in reality,the perception and control algorithms of each intersection are different,and the analysis and defense of signal attack are not universal.Therefore,this paper aims to explore a common signal attack analysis method and defense method,so as to provide more comprehensive support for the signal safety of intersections.These objectives are achieved through the following three studies.(1)The intersection adaptive control model and state table attack model based on reinforcement learning are established:the real-time state of vehicles in the lane is the state set,the phase is the action set,and the total waiting time of vehicles at the intersection is the reward.Taking advantage of the interactive learning between reinforcement learning model and environment,the best signal control scheme is obtained.The model-free control with both intensity learning makes it impossible for attackers to attack the defects of specific control logic.Meanwhile,a coordinated state table tampering attack between different lanes is designed to analyze the influence of signal attack on signal control more objectively.(2)Establish a fake trajectory injection attack model and detection model:a more realistic fake trajectory injection attack model is established.The attack method is to inject the network connected vehicle with fake parking at the end of the queue,and ensure that the network connected vehicle trajectory meets the vehicle dynamics constraints and has a certain attack target;In addition,mining the time-space characteristics of queued traffic flow,using the relative aggregation wave velocity of queued vehicles at the intersection as the similarity distance,a forged trajectory detection model based on hierarchical clustering is established,which provides a solution for the defense of intersection signal attack.(3)Build Sumo-Tensorflow joint simulation platform to simulate the intersection its environment,verify the performance of reinforcement learning control model,and evaluate the impact of signal attack on signal control model.The experimental results show that when attacking one lane,the delay is only increased by 7%,but when attacking two lanes,the maximum delay can be increased by 253%.The VISSIM-Sklearn joint simulation platform is established to realize the forged trajectory injection attack,and the abnormal trajectory detection model is verified and evaluated.In the case of different attack targets and permeability,the detection rate is the highest is 95%and the lowest is 67%,which verifies the effectiveness of the detection model. |
