Research On False Data Attack Detection Of Smart Grid Based On Machine Learning

With the deep penetration of information and communication technologies in modern power systems,power systems and information networks have gradually merged into Smart Grid.The development of smart grid has improved the production efficiency,management level and consumer participation of enterprises,but at the same time,it’s inevitable to face the security problem of information network.Information security accidents may even pass through the coupling relationship to the power grid to destroy the security and stability of the power systems.As a new type of network attack,the false data injection attack tampering with the measurement data collected by the SCADA systems,and obtaining the wrong state variables of the real-time state information of the power system through the state estimation,thereby disturbing the power system for power flow optimization,energy scheduling and emergency analysis.This paper studies how to detect false data injection attacks.The main research contents are as follows:Firstly,the impact of false data attacks on the actual operation process of power systems and specific equipment systems is studied.From the theoretical level,how the false data injection attacks pass the bad data detection test of state estimation is analyzed..Secondly,the principle of traditional false data injection attack detection method is studied.According to its shortcomings and the detection method of matrix transformation can’t be applied to large-scale power systems,statistical methods are used to analyze data changes to construct detection methods.After preliminary experiments,it is feasible but there are also deficiencies.Finally,the machine learning method is used to strengthen the statistical-based detection method.After the comparative analysis,the neural network is used to construct.Based on the optimization of learning efficiency,the detection method based on Extreme Learning Machines(ELM)is further proposed to improve the learning efficiency.And the detection method based on Random Vector Functional Link(RVFL)is also proposed to improve the detection accuracy and learning efficiency with the optimization of the neural network architecture.At last,applying the IEEE 14-node test system to obtain the system state estimation value and the measurement data by using the real-time load data of the New York independent system operator.The proposed attack detection method is tested in many aspects.The experimental results show the effectiveness of the two detection methods,and provide direction for safe and stable Smart Grid.