Detection Localization And Recovery Of False Data Injection Attacks On Power Grids Based On Neural Networks

The traditional power grid has essentially realized the transformation to a smart grid as a result of the advancement of advanced information and communication technology.An important feature of smart grid is the interaction of the cyber system and the physical system,resulting in a highly coupled power information and physical system,which exposes smart grid to more serious information security threats than traditional grid.False Data Injection Attack(FDIA),a new form of stealthy attack,is critical for maintaining the safe and stable operation of power systems by detecting and locating it quickly and accurately,as well as recovering normal data.The thesis uses FDIA on the power grid as the research object and primarily performs the following tasks:To begin,an FDIA base data set satisfying the actual scenario is built based on an AC power system with partially known network information in order to weaken the reliance on the complete network information of the power system in the FDIA construction process and further improve the concealment of FDIA construction;Then,a comprehensive set of detection,localization,and data recovery schemes based on neural network algorithm are suggested in order to effectively counter the threat of FDIA.In order to detect high-dimensional data accurately and quickly,the FDIA detection and localization scheme introduces a self-attention mechanism based on deep convolutional neural networks.In order to effectively recover the original data,the data recovery method is based on the correlation of the measurement data itself.It trains an autoencoder-long and short term memory neural network using historical measurement data,and then generates corrected data using the same distribution pattern.Finally,to improve the scheme’s adaptability,the FDIA construction,detection,localization,and recovery algorithm is subjected to systematic experimental validation using the IEEE 14 and 118 systems.The results show that the proposed complete set of algorithms performs admirably and can be used to effectively defend the power grid against stealthy FDIAs.