Classification of False Data Injection Attacks in Smart Grid using Multi Label Deep Learning

The traditional electricity structure is displaced by a new physical structure brought about by the smart grid, which is brought about by the creative use of information that leads to the building of a new structure. When applied to a sector of the economy like this one, the dissemination of bad information has the potential to have substantial repercussions. The conventional residual methods that are used in fake data identification will never be in a position to identify accurate FDI attacks. The linearized power system model DC has served as the foundation for the vast bulk of the research that has been conducted on the subject of the detection of FDI attacks. The AC model is not the source of any of the assaults that are uncovered by this study. In addition, the overwhelming majority of the currently available DIA identification algorithms focus on FDIA, despite the fact that significant injection site data cannot be gathered. In light of the ongoing advancements in deep learning, we present a Deep Learning based Locational Detection approach as a means of continually identifying the various parts of FDIA. Solver Gap Happiness is a False Data Detector (FDD) that was developed in the realm of software engineering. It makes use of a Recurrent Neural Network (RNN). The FDD has been around long enough to be able to spot false information. The subsequent RNN is employed here as a multi-label classifier in order to analyse the irregularity and co-occurrence dependence of power flow computations which may have been caused by potential assaults. Due to the fact that they are "model-free," the suggested design does not include any prior statistical assumptions. In addition to this, it is "cost-accommodating," since it does not change the existing FDD structure and the identification method only takes a few microseconds on a typical home computer. Through our extensive work in the IEEE 14, 30, 57, and 118 bus systems, we have shown that ANN-MLP, SVM-RBF, CNN and RNN are capable of performing locational detection in a variety of environments with varying levels of noise and attack. In addition to this, the multi-name classification approach that was applied manages to effectively increase the accuracy of the presence identification.