Research On False Data And Prediction Based Detection Scheme In Power System

With the rapid development of Internet of Things technologies,modern power systems have become complex cyber-physical systems.A large number of smart devices have promoted efficient generation,transmission and distribution in the smart grid.State estimation(SE)is one of fundamental components in smart grid that evaluates the operation state of a grid by using a set of sensor measurements and grid topologies.A major issue is the authenticity of the measurements collected by the sensors.One of the various ways to influence measurements is called false data injection attack,which is a method of injecting false data into the grid from the perspective of an attacker.The method achieves the effect of interference state estimation by premeditated tampering of the measurement of multiple measuring instruments,and can bypass the residual-based data detection module,thereby jeopardizing the safe and stable operation of the system.This thesis studies the problem of false data injection attack in the power grid.The specific contents include:1.This thesis introduces the basic knowledge of data and state estimation in the power system,and the commonly used residual-based detection and identification methods.The false data injection attack is introduced and data prediction-based method—PDL is adopted for the attack.The linear vector autoregressive model is used in PDL to fit the data in the power system,and then according to the distribution of the difference between predicted value and observed value,detection of abnormal data is performed at detection point.The results show that the detection rate is as high as 90% or more.2.Considering that the data mostly contains both linear and nonlinear components,the PDL scheme is improved.The improved scheme uses residual recurrent neural network to predict the data in the grid and fits the Weibull distribution according to the sum of squared errors of the predicted value to determine the threshold and detect abnormal data.The residual recurrent neural network consists of two phases: the first phase uses a linear vector autoregressive model.The second phase predicts the error of the first stage using the nonlinear recurrent neural network to obtain the final predicted value.As can be seen from the results,the detection accuracy is improved compared to the pure linear prediction scheme.3.In the concept of false data injection attack,it is assumed that the attacker knows the complete knowledge of the topology of power grid,but this is not practical.A new attack construction method is introduced for this problem.The method firstly performs principal component analysis on the observed values,and then constructs a specific attack vector.Lastly,verifying the detection performance of the data prediction based method for such a new attack.