False Data Injection Attacks And Defense Methods For The Distributed Estimation In Large-scale Networks

With the rapid development of information technology and the continuous expansion of network scale,the architecture of large-scale distributed network has attracted wide attention in recent years.The distributed estimation is one of the most important applications in large-scale distributed networks.And the security of distributed estimation is one of the popular focuses in the research field of network security.Due to the openness of network,attackers can launch the false data injection attack(FDIA)during the implementation of distributed estimation.In this dissertation,the FDIAs and defense methods for the distributed estimation in large-scale networks were investigated.We focused on four distributed estimation scenarios in two different network architectures,where the potential FDIA risks,the corresponding FDIA detection and defense methods were investigated.The main contributions of this dissertation are summarized as follows.(1)For the distributed network parameter estimation,we investigated the FDIAs against the distributed smart grid state estimation and network loss tomography,as well as the corresponding FDIA defense methods.In order to resolve the rank deficiency problem of the residual covariance for the real-time FDIA detection mechanisms in distributed smart grid state estimation,a residual prewhitening method was proposed to compress the dimention of residual and make the probability density function of the prewhiten residual become calculable for the fusion center.For the distributed network loss tomography,the bad data processing(BDP)mechanism was utilized to evaluate the abnormality of measurements,and the abnormal measurements can be removed from the measurement model,in order to mitigate the impact of abnormal measurements on the performance of network loss tomography and improve the reliability.In addition,the security breach caused by irredundant measurements was utilized to design a stealthy gray-hole attack strategy,and a corresponding attack defense method was proposed.By eliminating all of the irredundant measurements in the measurement model,the stealthiness of gray-hole attack can be destructed,and the BDP mechanisum can indentify and remove the compromised measurements.The research showed that the proposed FDIA defense method can resist the stealthy gray-hole attack effectively,as well as decrease the estimation error of network loss tomography.(2)The potential FDIA problem and the FDIA defense methods in distributed network topology inference were investigated.First,from the perspective of an attacker,an FDIA strategy was proposed,where malicious nodes inject some false data in the end-to-end transmission delay measurements by postponing the forward of packets,in order to cause an incorrect result of network topology inference.The research showed that attackers can launch five categories of FDIA by adjusting the attack parameters,which can decrease the correct probability of network topology inference.In order to defend against the FDIA,an FDIA detection method was proposed based on the Pearson goodness-of-fit test technique.And the theoretical performance of the proposed FDIA detection method was derived.The research showed that the proposed FDIA detection method can achieve a high detection probability of the FDIA,as well as maintain a low false-alarm probability.(3)The acquirement of information of weight matrix was invsetigated for the FDIA detection mechanisms in consensus-based distributed estimation.And the effect of information of weight matrix on the stealthy FDIA strategies and the defense methods were studied.In the circumstance that multiple malicious nodes in the network cooperate with each other,a cooperative estimation method of the weight matrix was first proposed,and then the information of weight matrix was utilized to construct a stealthy FDIA strategy.By injecting some pre-designed false data into the information shared with neighboring nodes,malicious nodes can make the estimation result at each normal node converge to a false value.In order to defend against the FDIA,a real-time FDIA detection mechanism was proposed,where a critical step is designing a distributed estimation method of the eigenvalues of weight matrix based on the spatial-difference-based statistics.The estimated eigenvalues of weight matrix were further utilized to construct a real-time statistic for each pair of neighboring nodes.Based on the proposed real-time FDIA detection mechanism,resilient weights are designed with the maximum-degree protocol and Metropolis protocol.The research showed that the proposed real-time FDIA detection mechanism can achieve a high attack detection probability,as well as maintain a low false-alarm probability.(4)The process methods of high-power system noise and the FDIA defense methods in the consensus + innovation distributed filtering were investigated.First,the Cauchy-Schwarz inequality was utilized to derive a modified upper bound of estimation error covariance.It was proved that the modified upper bound of estimation error covariance is tighter than existing upper bounds.In order to defend against the FDIAs launched by malicious nodes in the network,an FDIA detection method was proposed based on the local Kalman filtering,where the benchmarks of measurements and state vectors were constructed with the results of local Kalman filtering,and the FDIAs can be detected with the normalized deviations between the received data and the corresponding benchmarks.Next,a sub-optimal design protocol for resilient weights was presented.The convergence property of estimation error of the consensus+ innovation distributed filtering under the presented sub-optimal weights was proved.The research results in this dissertation can provide theoretical guidance and technical support for the future applications of distributed estimation in large-scale networks.