Cyber Attack Detection Process in Sensor of DC Micro-Grids Under Electric Vehicle Based on Hilbert–Huang Transform and Deep Learning

In this article, a new procedure is proposed on the basis of Hilbert-Huang Transform and deep learning for cyber-attacks detection in direct current (DC) micro-grids (MGs) as well as detection of the attacks in distributed generation (DG) units and its sensors. An advanced elective group deep learning method with Krill Herd Optimization (KHO) algorithm is proposed. At first, Hilbert-Huang Transform is used with the aim of extracting the signals feature and next these features are applied as the multiple deep input basis models are made with the aim of capturing automatically sentient traits from raw fluctuation signals. At third, to make sure the variety of the basis patterns, linear decoder, denoising autoencoder and sparse autoencoder are applied to make various deep autoencoders, respectively. Further, Bootstrap is applied with the aim of designing separate educational data subsets for any base model. Fourth, for implementing selective ensemble learning, a combination strategy of enhanced weighted voting (EWV) with class-particular thresholds is studied. Eventually, KHO algorithm is applied with the aim of adaptive selecting the optimal class-specific thresholds. In the offered tactic, firstly, a DC micro-grid is functioned and controlled with the lack of any false data injection attacks (FDIAs) to collect adequate information within the usual operation needed for the educating of deep learning networks. It is noteworthy that, in the procedure of datum production, load variable is also determined with the aim of having distinctive datasets for cyber-attack scenarios and load variables. Also, to provide more realistic method, the smart plug-in electric vehicle is also considered in the model. Outcomes of Simulation in various scenarios are applied with the aim of verifying the benefit of the offered procedure. The outcomes propose that the offered procedure is able to more accurate and robust know various type of false data injection attack over than 93.76% accuracy detection of true rate.