Research On Cascading Failure Models And False Data Injection Attacks Of Cyber Physical Power System

As a key part of the Energy Internet, the modern power system has been greatly developed to the cyber physical power system (CPPS), which is a deep integration of the information network (i.e. communication system, monitoring and control system) and power grid (i.e. electric power system). Due to the deep integration of information network and power grid, the electric power system has many advantages, such as flexibility, autonomy and high efficiency, and can better serve the Energy Internet. But the CPPS also faces many challenges, including the faultiness of fundamental theories and system models, vulnerability, security and standardization, etc.Among the challenges, on the one hand the deep integration of information network and power grid increases the structural vulnerability of the power system. When failures occur in the information network or power grid, the coupling between the information network and power grid will cause failures to be mutual conversion of them and cascading failures occur more easily. One the other hand, the deep integration of information network and power grid causes the power system to face risks of cyber attacks. As many potential security holes, defects and failures exist in softwares of the information network, there are many entries of cyber attacks without influences of many factors, such as weather, geography and time, especially cyber attacks are more stealthy. As a new cyber attack which takes advantage of the flaws of classic state estimation, the false data injection attacks can cause serious damages to the safe and stable operation of power system and even trigger electric power system blackouts. Therefore it is of great theoretical and practical significance to study cascading failures and false data injection attacks of CPPS.Based on the complex network theory and power system state estimation theory, this dissertation focuses on cascading failure models and false data injection attacks. The main contents are presented as follows.(1) Considering the community structure of the information network and power grid, three coupling preferences are proposed, i.e. random coupling in communities (RCIC), assortative coupling in communities (ACIC) and assortative coupling with communities (ACWC). These coupling preferences can reflect the effects of the community structure on node couplings. The information network and power grid are abstracted into interdependent modular scale-free networks and the cascading failure model of interdependent modular scale-free networks is built, considering the loads and capacities of nodes. The results show that decreasing modular strength of the information network and power grid can enhance the robustness of the power system. The loss is larger caused by hub attacks than inner attacks. The ACIC is more beneficial to the robustness compared with RCIC or ACWC. The ACIC also can prevent the cascading failures propagating globally and the cascading failures propagate mainly in a local community where the initial failure occurs.(2) As communication optical cables are damaged more easily than information nodes, a cascading failure model is proposed, which considers the edge failures of the information network from the perspective of deep integration of the information network and power grid. By analyzing the coupling of the information network and power grid, the coupling strength q is introduced to reflect the impacts of the power grid on the information network. The information network influences the power grid through dispatch. The proposed cascading failure model combines the fast dynamics of the OPA model and hidden failures. The results show that as the scale of the edge failures increases, the strength that communication optical cables impact the cascading failures of the power grid varies from strength to weakness, eventually to stability. Moreover, topologies of the information network influence the cascading failures of the power grid strongly and a broader degree distribution of information network increases the vulnerability of the power grid. The average load loss is larger when the dispatching center is set on the higher degree node. When the dynamics (for example, route) on the information network are neglected, the coupling strength q has no distinct impacts on cascading failures in the power grid.(3) As there may be errors and incompleteness in branch susceptances that attackers master or bad data in measures, a method of constructing false data injection attack vectors is proposed based on the parameter estimation. The Lagrange multipliers and augmented state estimation method are used to identify bad data and estimate unknown branch susceptances. Then the false data injection attack model aiming to attack the measurement k is transformed into BP (Basis-Pursuit) model by the convex relaxation. Finally, the suboptimum attack vector is quickly solved with ADMM (Alternating Direction Method of Multipliers). The results show that it is more efficient with ADMM compared with classic LP (Linear Programming) based on BP model. Moreover, the successful rate of the false data injection attacks is relatively high and the attack cost does not increase significantly with ADMM.(4) Considering that the classic countermeasures of protecting measures do not balance the relationship between the part and the whole of the power system, a new countermeasure is proposed base on the index Vk and Rk.The index Vk and Rk measure the ability to defense false data injection attacks from local and global perspective respectively. As the defensive model based on the index Vk and Rk is a NP-hard problem, the method based on the importance index Ik is proposed to solve the model. The results show that when the percentage p of protecting measurements is relatively small, the index Vk decreases and the index Rk increases, meaning that the index Vk and Rk can reflect the protective effects. Moreover, the countermeasure based on index Vk and Rk can increase the attack cost and hardness.