Deep Learning-Based Sybil Attack Detection Methods In VANET

With the steady increase in the number of vehicles and the growing societal concern for road safety,Vehicular Ad hoc Networks(VANET)play a key role in Intelligent Transportation Systems.However,the mobility and dynamic topology of VANET present significant security challenges,especially those posed by Sybil attacks.These attacks,by creating a multitude of fake identities,threaten other users’ access to the network,interfere with communication between nodes,and mislead routing choices,thereby severely compromising the security of VANET.To address this challenge,two main tasks were undertaken in this paper:(1)This paper designs a detection model for Sybil attacks in VANET based on LSTM-BiGAN.The model uses the basic safety messages(BSM)of normal vehicles as training samples and is capable of reconstructing normal data after learning.In the detection phase,the model reconstructs the received BSM data and calculates the anomaly scores between the original and reconstructed BSM to distinguish between Sybil nodes and normal nodes,achieving the detection of Sybil attacks.The model was trained and validated using the public dataset Ve Re Mi,and the results show that compared to other detection methods based on machine learning and deep learning,this model has a significant advantage in detection accuracy.(2)In response to the limitations of the Sybil attack detection model on individual vehicle nodes,namely,the inability to acquire all information from all vehicle domain nodes,this paper proposes a verification algorithm.The algorithm models vehicle travel information as a motion matrix and uses a spectral clustering algorithm to cluster the nodes in each small vehicle domain.The clustering results are combined with the output of the detection model and each node’s trustworthiness is determined using a trustworthiness calculation formula.In addition,a Bayesian method is used to dynamically classify newly added vehicle nodes and the results are fed back to two key parameters in the detection model for timely model updates.To evaluate this verification algorithm,this paper uses the F2 MD simulation framework to simulate two types of Sybil attacks and test the necessity of the verification model.The results show that this verification model effectively reduces the risk of false alarms and missed reports,enhancing the precision and robustness of Sybil attack detection.In summary,the Sybil attack detection model and Sybil node verification algorithm proposed in this paper,both based on LSTM-BiGAN,together constitute a solution for addressing Sybil attacks in VANETs.Experimental results demonstrate that both methods significantly enhance the security of VANETs,mitigating the threats posed by Sybil attacks.