摘要
Without the known state equation, a new state estimation strategy is designed to be against malicious attacks for cyber physical systems. Inspired by the idea of data reconstruction, the compressive sensing （CS） is applied to reconstruction of residual measurements after the detection and identification scheme based on the Markov graph of the system state, which increases the resilience of state estimation strategy against deception attacks. First, the observability analysis is introduced to decide the triggering time of the measurement reconstruction and the damage level from attacks. In particular, the dictionary learning is proposed to form the over-completed dictionary by K-singular value decomposition （K-SVD）, which is produced adaptively according to the characteristics of the measurement data. In addition, due to the irregularity of residual measurements, a sampling matrix is designed as the measurement matrix. Finally, the simulation experiments are performed on 6-bus power system. Results show that the reconstruction of measurements is completed well by the proposed reconstruction method, and the corresponding effects are better than reconstruction scheme based on the joint dictionary and the traditional Gauss or Bernoulli random matrix respectively. Especially, when only 29% available clean measurements are left, performance of the proposed strategy is still extraordinary, which reflects generality for five kinds of recovery algorithms.
Without the known state equation, a new state estimation strategy is designed to be against malicious attacks for cyber physical systems. Inspired by the idea of data reconstruction, the compressive sensing （CS） is applied to reconstruction of residual measurements after the detection and identification scheme based on the Markov graph of the system state, which increases the resilience of state estimation strategy against deception attacks. First, the observability analysis is introduced to decide the triggering time of the measurement reconstruction and the damage level from attacks. In particular, the dictionary learning is proposed to form the over-completed dictionary by K-singular value decomposition （K-SVD）, which is produced adaptively according to the characteristics of the measurement data. In addition, due to the irregularity of residual measurements, a sampling matrix is designed as the measurement matrix. Finally, the simulation experiments are performed on 6-bus power system. Results show that the reconstruction of measurements is completed well by the proposed reconstruction method, and the corresponding effects are better than reconstruction scheme based on the joint dictionary and the traditional Gauss or Bernoulli random matrix respectively. Especially, when only 29% available clean measurements are left, performance of the proposed strategy is still extraordinary, which reflects generality for five kinds of recovery algorithms.
基金
This work was supported by the Natural Science Foundation of China （NSFC）-Guangdong Joint Foundation Key Project （No. U1401253）, the NSFC （Nos. 61573153, 616721 74）, the Foundation of Guangdong Provincial Science and Technology Projects （No. 2013B010401001 ）, the Fundamental Research Funds for the Central Universities （No. 2015ZZ099）, the Guangzhou Science and Technology Plan Project （No. 201510010132）, the Maoming Science and Technology Plan Project （No. MM201 7000004）, and the National Natural Science Foundation of Guangdong Province （No. 2016A030313510）.
