Controller area network (CAN) is the most commonly used bus technology for In-vehicle network and uses multicast communication without corresponding security measures. Therefore, the message data field is vulnerable to tampering and other attacks. Recent machine learning-based intrusion detection methods for CAN bus messages only use the information contained in the message data field and do not take into account the contribution made by the neighboring information of CAN bus messages. In addition, previous models considered the data domain information of CAN bus messages as separate features and did not consider the unique weight of each feature, as well as the second-order interaction information between features. Therefore, we propose a novel intrusion detection model, The Hybrid Similar Neighborhood Robust Factorization Machine Model (HSNRFM), for detecting anomalies in CAN bus messages to address the shortcomings and problems of the previous models. To be able to incorporate the contribution of similar neighborhood information and learn the unique weight parameters of each feature in the model decision process, as well as additional second-order interaction information between features, the HSNRFM model solves the above problem using a similarity calculation method and a factorization machine model. Comprehensive experimental results are compared on real vehicle datasets. The HSNRFM model has AUC values of 0.9216 and 0.901 and AUPR values of 0.9194 and 0.9018 on two real datasets, respectively. And the results show that our proposed HSNRFM model has excellent detection efficiency for intrusion detection of CAN bus messages.
As a state-of-the-art distributed learning approach, federated learning has gained much popularity in connected and autonomous vehicles (CAVs). In federated learning, models are trained locally, and only model parameters instead of raw data are exchanged to aggregate a global model. Compared with traditional learning approaches, the enhanced privacy protection and relieved network bandwidth provided by federated learning make it more favorable in CAVs. On the other hand, poisoning attack, which can break the integrity of the trained model by injecting crafted perturbations to the training samples, has become a major threat to deep learning in recent years. It has been shown that the distributed nature of federated learning makes it more vulnerable to poisoning attacks. In view of this situation, the strategies and attacking methods of the adversaries are worth studying. In this paper, two novel optimization-based black-box and clean-label data poisoning attacking methods are proposed. Poisoning perturbations are generated using particle swarm optimization hybrid with simulated annealing and genetic algorithm respectively. The attacking methods are evaluated by experiments conducted on the example of traffic sign recognition system on CAVs, and the results show that the prediction accuracy of the global model is significantly downgraded even with a small portion of poisoned data using the proposed methods.
10 October 2022 in IET Intelligent Transport Systems
This paper proposes an adaptive sliding mode control based on an auxiliary distance sensor for nonlinear platoons over vehicular ad‐hoc networks under Denial‐of‐Services (DoS) attacks and external disturbances. DoS attacks can cause packet dropping for the wireless network and will lead to the collision or degradation to adaptive cruise control of platoons. To solve this problem, an adaptive Radial Basis Function (RBF) sliding mode control method is proposed for the attacked platoons. RBF neural networks are used to approximate the unknown nonlinear part and external disturbances contained in the vehicle models. Then the adaptive sliding mode controllers are designed to achieve the control goals: the spacing error converges to zero, and the platoon maintains the string stability. The system's internal stability and string stability are proved. Numerical examples are applied to show the effectiveness of the proposed method. Compared with the existing method, the spacing error is decreased from 5.0726, 6.4637 to 0.0264 for secure distributed adaptive platooning control in PF topology and LPF topology, and the average distance is increased from 7.0580 m, 5.7442 m to 11.8498 m for secure distributed adaptive platooning control in PF topology and LPF topology to maintain a safe distance under the situation of DoS attacks.
16 January 2023 in Optimal Control Applications and Methods