|
Abstract
|
Smart grids have advanced communication technologies that make them vulnerable to cyber-attacks. This paper
investigates a new method called a supervised convolutional neural network (CNN)-based system state estimator,
which consists of two systems: a data validation model and a fault detection model. In this research, heat graphs
are used as a tool for visualizing numerical data and a method for preprocessing information in training the CNN
model. On the other hand, one emerging tool in artificial intelligence is generative adversarial network (GAN),
which is a deep learning method that can bypass intrusion detection systems by generating fake examples. In
response, this paper uses both common examples and examples generated by GAN for training and evaluation of
the presented system. The proposed method considers the role of fake data injection (FDI) in smart networks with
the aim of increasing the accuracy of the proposed model and then focuses on the impact of combined FDI and
denial of service (DoS) attacks on smart networks. In this paper, two complex cyber-attack scenarios are
investigated. The proposed model is capable of effectively detecting both attacks, as evidenced by the simulation
results for varying attack intensities, which achieved a validation accuracy of 99.51 in scenario I and 99.32,
99.59, and 99.50 in scenario II. Moreover, the results indicate that using the examples generated by GAN helps
the data validation model to increase its accuracy against cyber-attacks and the fault detection model quickly
identifies the desired fault and notifies the system operators.
|