Research On Detection Of False Data Injection Attack In Cyber Physical Power System

The growing demand for reliable energy and the rapid development of power information technology have promoted the progress of modern power system,which has gradually transformed from traditional power system to the cyber physical power system(CPPS)with deep integration of information system and physical system.However,the frequent network security problems make the CPPS unable to operate safely and reliably.Among them,the false data injection attack(FDIA)is considered to be one of the most threatening network security problems for the safe operation of the system by virtue of the feature that they can evade traditional bad data detection.In view of the fact that the existing FDIA detection methods rely on specific estimation methods and system knowledge,which have the disadvantages of small amount of processing data,low efficiency,and limited application scenarios to traditional power systems,resulting in low feasibility in the CPPS.Given the above problems,this thesis uses the advantages of artificial intelligence algorithm in dealing with detection problems and proposes new FDIA detection methods.The main work and contributions are as follows:First of all,the data samples are constructed.Starting from the engineering practice,the FDIA initiated by the attacker when grasping part of the information is used as the attack vector,and the Matpower simulation tool kit is used to generate the positive and negative data sample sets for the detection of IEEE14-bus,IEEE39-bus and IEEE118-bus three different scale standard test systems,which lays the data foundation for the construction and training of subsequent detection models.Secondly,the data samples are processed.Because of the high dimension and strong noise of the CPPS measurement data,the construction and training effect of the direct detection model is not good.Therefore,this thesis uses the uniform manifold approximation and projection algorithm to process the original measurement data,captures the global structure and local structure,and generates a low-dimensional and high-quality experimental data set,which provides a strong guarantee for building a real-time and efficient detection model.Finally,the detection model is constructed and the simulation experiment is carried out.To solve the problem that the traditional detection method cannot effectively detect the FDIA in the CPPS,this thesis selects support vector machine as the weaker classifier,cascades the weaker classifier by gentle adaboost algorithm,and iteratively forms a strong classifier,constructs a detection model based on SVM-GAB.The simulation results show that the detection model has good detection performance and economy in small-scale systems such as IEEE14-bus,and improves the detection accuracy while ensuring the detection efficiency.However,when the system scale expands and the complexity increases,the above detection model is no longer applicable.Therefore,based on the deep neural network that is good at dealing with big data problems,this thesis combines the minimal gated unit with the full connection layer to construct the detection network.Using the spatial and temporal data characteristics of continuous system state,the FDIA data and normal data in the system are distinguished,and a detection model based on DNN is constructed.The simulation results show that the model effectively improves the detection accuracy and detection efficiency in large-scale systems such as IEEE118-bus.In summary,considering the detection performance and cost,the detection model based on SVM-GAB is adopted in small-scale CPPS,while the detection model based on DNN is adopted in large-scale CPPS,which can better realize the real-time,efficient and economic detection of FDIA,and thus improve the security and stability of the CPPS.