Anomaly Detection Of Smart Grids Under False Data Injection Attack

Smart grid has become one of the most important infrastructures in today’s society.Compared with traditional power system,the distinguishing feature of smart grids is increasingly tight coupling and interaction of information and physical elements.And the basic function of energy management system,like optimal control and economic dispatch,is achieved through the information interaction in the information layer.However,as the grid becomes more and more intelligent,it also becomes more and more vulnerable.A large number of unprotected sensing devices and increasingly complex connection networks give attackers the opportunity.Existing research shows that attackers can use the leak in the information layer to launch attacks and induce the system to make wrong decisions,which brings very serious physical,economic and social consequences.One of the most typical attacks is false data injection attack(FDI).It has attracted widespread attention due to the ever-changing attack targets and methods,as well as a certain degree of concealment.The research on false data injection attacks will benefit the protectors of smart grids in order to improve the performance of the original protection system.Thus,it has very important theoretical and practical significance for the safe operation of smart grids.The thesis focuses on a class of false data injection attacks which is based on continuous load redistribution(CLR).A continuous load redistribution attack model is constructed.An anomaly detection algorithm and a coder-decoder based amplification algorithm for CLR based FDI attacks are designed to achieve system safety.The main contributions of this thesis are summarized as follows:Firstly,based on the analysis of the mechanism of false data injection attacks and the optimal economic dispatching strategy of the smart grid energy management,a continuous load redistribution attack model is constructed.The attack target is to make the energy management center doing load shedding according to the impacted data.The impact of continuous load redistribution attack model on the power system is analyzed with emphasis on the limited attack resources.The simulation verifies the feasibility of the continuous load redistribution attack model.Secondly,an anomaly detection system using multidimensional normal test statistical characteristics and a supervised ensemble tree model is designed.This method takes into account the continuous and small amplitude characteristics of continuous load redistribution attacks,calculates the multidimensional skewness and kurtosis statistical characteristics,and trains a multi-classification model based on ensemble tree model in the smart grid operation simulation dataset.The training results show that the designed anomaly detection system can effectively identify the target bus who has been attacked.Finally,due to the relatively small magnitude in the initial stage of the attack,the accuracy of the detection by the anomaly detection system given above is relatively low.Thus,an encoder-decoder-based channel encryption amplification algorithm is proposed.The algorithm can amplify the attack signal in the channel,which is beneficial to the detection of the anomaly detection system.At the same time,the coder-decoder has a certain encryption ability and cannot be easily attacked.Simulation test is performed in the IEEE-14 bus test system to verify the effectiveness of the algorithm.