Research On Distributed Secure Estimation Algorithm Based On Local Outlier Factor

Distributed estimation on wireless sensor networks(WSNs)refers to the collaborative estimation process of parameters of interest from a noisy data set by interconnected nodes without a fusion center.In recent years,this technique has received a lot of attention,benefiting from the rapid development of the internet of things and the industrial internet.However,the complex working environment and the collaborative nature of the nodes make the network vulnerable to intrusion by attackers in an adversarial network environment,which can lead to incalculable damage.Common attacks in adversarial networks include false data injection attack(FDIA)and byzantine attacks(BAs).This thesis,therefore,focuses on the distributed secure estimation problem in multiple adversarial WSNs in the presence of FDIAs and BAs and proposes corresponding solution algorithms based on local outlier factor(LOF)strategy.First,a secure low traffic diffusion least mean square(SLT-DLMS)algorithm is proposed for the intrusion of attackers in single-task networks.The algorithm exploits the feature that the parameters to be estimated are the same in a single-task network and designs an attack detection module based on the local density of the nodes’ intermediate estimates and the LOF strategy to detect malicious nodes that are under attack.Meanwhile,during the information fusion process,the SLT-DLMS algorithm replaces the intermediate estimate of the malicious node with the intermediate estimate with the highest local density,resulting in the ideal parameter to be estimated.Theory and experiments demonstrate that the SLT-DLMS algorithm achieves secure estimation while reducing communication by half compared to classical two-system distributed secure estimation algorithms on single-task networks.Secondly,for the problem of FDIA and BAs in multitask networks where cluster information is known,using inter-task similarity and the LOF strategies,this thesis proposes the average-fusion-strategy-based multitask secure DLMS(AFS-MSDLMS)algorithm and LOF-fusion-strategy-based multitask secure DLMS(LOF-MSDLMS)algorithms.The difference between the two algorithms is that different fusion strategies are used in the information fusion step.Both the AFS-MSDLMS and LOF-MSDLMS algorithms have been theoretically and experimentally proven to resist FDIA and BAs and derive correct estimates.While the LOF-MSDLMS algorithm can dynamically adjust the weights of the nodes in the information fusion process according to the local density of the intermediate estimates of the secure nodes,so this algorithm has an advantage over the former in combating symbol flipping attacks in BAs.Finally,this thesis proposes a two-system-clustering-based secure DLMS(TCS-DLMS)algorithm for the problem of distributed secure estimation in networks with unknown a priori information.The algorithm consists of a non-cooperative LMS(NC-LMS)subsystem and a cooperative LMS(C-LMS)subsystem.The NC-LMS subsystem detects the secure homogeneous network topology of any node using the LOF strategy,while the C-LMS subsystem performs information fusion based on the topology detected and thus estimates the parameters to be estimated.The TCS-DLMS algorithm is theoretically and experimentally proven to be robust to FDIAs and BAs in networks with unknown a priori information,and can accurately estimate the parameters of interest in an adversarial network environment.