Search

This paper considers the event-triggered consensus problem for multi-agent systems (MASs) under intermittently random denial-of-service (IRDoS) attacks. The IRDoS attack compromises communication networks when the attacker is at active periods. Then, the distributed event-triggered controller is designed to achieve the consensus objective, and each agent's communications towards neighbors are interrupted by the IRDoS attack with a certain probability when the triggered instants are at active periods of attackers. Furthermore, the sufficient conditions on the attack duration and successful probability, under which the designed control protocol guarantees secure consensus of MASs, are proposed. Finally, the effectiveness of the distributed event-triggered controller under a class of IRDoS attack is illustrated with a numerical example. [ABSTRACT FROM AUTHOR]

Highlights • A stochastic coding scheme is proposed to generate covariance difference of the compromised data under replay attack, which removes the performance degradations in existing works. • A residual-based detector is designed to detect the covariance changes caused by the coding scheme subject to the false alarm rate (FAR) and attack detection rate (ADR) limitations. Then, an output-based detector is proposed, which effectively reduce the design complexity of the residual-based detector. • Different from traditional detector, the above detectors apply a novel form to remove the non-convex constrain conditions in design optimal problems. Abstract In this paper, the security problems in cyber-physical systems (CPSs) against replay attack are considered. With replay attacks, attacker records and covers the transmitted data between the senders and receivers of the sensors. In order to achieve the detection objective for malicious replay attacks, the stochastic coding scheme is proposed to make the CPSs generate covariance differences between the normal and compromised data. Different from the existing results, this method detects the replay attack without sacrificing any system performances in normal systems. Based on the coding scheme, two types of detectors are further designed to detect the covariance changes in residual and output, respectively. Moreover, the output-based detector has more low-computing solution process than the residual-based one. Finally, a practical example is proposed to demonstrate the superiority of the stochastic coding scheme. [ABSTRACT FROM AUTHOR]

Domain adaptation is one of the mainstream deep transfer learning strategies to deal with unsupervised fault diagnosis issues. Nevertheless, the existing domain adaptation based diagnosis approaches mainly focus on the alignment of source domain and target domain at global level, which may lead to the confusion between subdomains and finally restrict the improvement of target domain accuracy. In this work, the subdomain adaptation capsule network (SACNet) is developed, and the contributions of the proposed method mainly reflected in the following aspects. Firstly, in the proposed architecture, a novel distance metrics named local joint maximum mean discrepancy (LJMMD) is developed and embedded as part of the objective function. Utilizing the LJMMD, the joint distribution of feature and prediction space of target domain can be matched with that of source domain at subdomain level, which can avoid the limitations of global or single marginal distributions match. Secondly, the capsule network with vector-output is substituted for the classical fully connected adaption layer. Instead of matching specific values, the module value of the capsule vector is set as the object of domain adaption. Thus, the discrepancy between different domains can be admitted to a certain extent, which can endow the model with more excellent generalization ability. The experimental results indicate that, compared with other methods, the suggested LJMMD can achieve higher diagnostic accuracy in most transfer tasks. Besides, the diagnostic accuracy of the developed SACNet can be further improved by nearly 2% after adopting the capsule adaptation layer. [ABSTRACT FROM AUTHOR]