Defense Methodology Of Cyber-Physical Attacks Against Power System State Estimation

State estimation is the core application of energy management systems deployed in control centers.It takes the responsibility of accurately estimating the operating state of a power system using measurements collected all over the grid.Its results are both the input of other energy management applications and the reference of operators for decision-making,which stresses its significance in ensuring the stable and economic operation of power systems.With the emergence of new power appliances such as distributed generations and energy storage units,along with the progress of the grid structure,the coordination and interaction between different entities in power systems are becoming more frequent,which popularizes non-proprietary information and communication technologies in power dispatching automation systems.The cyber system and the physical system become tightly coupled and interactional,boosting traditional power systems to evolve into cyber-physical power systems.Introducing advanced information technologies not only improves the automation level of power regulation but also makes cyberattacks more impactive to physical power systems.Attackers are no longer satisfied with simply jeopardizing cyber systems and causing losses to cyber assets,but exploit cyber attacks to interfere with the normal operation of physical systems,launching the so-called “cyber-physical attacks”.In recent years,cyber-physical attacks against power system state estimation have attracted global attention both in academia and industry.This dissertation focuses on a series of in-depth research on the defense methodology against such attacks,with the main contents summarized as follows.A selection method of practical measurement protection schemes for preventing false data injection attacks is proposed.The proposed method first quantifies four indicators,including the risk exposure of cyber-physical attacks,the risk mitigation rate of protection measures,the cost of measurement protection,and the protection discount rate.It then uses the indicators to formalize the metric of return on investment to characterize the cost-effectiveness of protecting a set of measurements.The optimization problem maximizing the overall return on investment is transformed into a minimum Steiner tree problem and is solved based on a tree-pruning heuristic algorithm to find a set of measurement devices with the highest technical cost-effectiveness.Case studies validate the feasibility and efficiency of the proposed method.A secure communication protocol for protecting the confidentiality and integrity of power system dispatching and control communications is proposed.The protocol does not use any computationally intensive algorithm such as asymmetric cryptography,which makes it friendly to measurement devices with constrained computing capability.In addition,the proposed protocol sets up a synchronously rolling pseudo-random number generator in each node taking part in a secure session and creates an independent key for each message in the session based on a random number from the generator.This feature makes it more appropriate for communication scenarios like power dispatching considering the requirements of long session duration.Besides,the proposed protocol not only supports unicast but also multicast and broadcast communications,which further strengthens its application potential.Experiments based on colored Petri nets and cyber-physical co-simulation verifies the robustness,effectiveness,and execution efficiency of the protocol.A measurement manipulation detection method oriented to highly imbalanced training samples is proposed.The method adopts supervised learning to detect possible measurement manipulation attacks in the power system based on the distinct patterns between normal and abnormal measurements stored in the historical operational database.In order to prevent the prediction bias towards the majority category caused by the rare abnormal training samples,the method uses data rebalancing techniques to synthesize abnormal samples and uses ensemble learning to mitigate the high variance of detection results stemming from the randomness of the data rebalancing process.Test results show that the proposed method can achieve satisfactory detection performance when training with highly imbalanced historical measurements.A distributed semi-supervised measurement manipulation detection method based on deep autoencoders is proposed.The method comprises a set of distributed anomaly measurers,each of which is deployed on a transmission line to evaluate the anomaly degree of the phasor measurements collected from both sides of the line.Relative to detection frameworks with centralized architectures,the proposed method is more suitable for large-scale power systems.The application of semi-supervised learning significantly relieves the burden of manually labeling training samples and can avoid inadequate learning of abnormal patterns because of the insufficiency of manipulated measurements in practice.Compared with conventional machine learning models,it can be discovered from test results that the proposed deep learning-based method not only has superior detection accuracy but also has better generalization and antinoise capabilities.A privacy-preserving hierarchical state estimation framework for large-scale interconnected power systems is proposed.In order to prevent the sensitive operational data of some participants from being stolen by others during state estimation,the proposed framework adjusts the regular procedure of hierarchical state estimation and introduces the degree-2 thresholded Paillier homomorphic cryptosystem into the interactions of different participants.The changes allow the untrustworthy higher-level control center to carry out the coordination required for hierarchical state estimation without knowing the true values of the data reported by lower-level control centers.Experiments based on a leased cloud server show that the degradation of state estimation accuracy brought by the proposed framework is negligible.Besides,its computation efficiency meets the practical requirement,rendering its application potential in large-scale interconnected power systems.