A Deep Learning Based Cyber Attack Detection Scheme in DC Microgrid Systems

Abstract

In this article, a dual deep neural network (DDNN) based cyber-attack detection and correction method for direct current microgrids (DCMG) are proposed. DCMG are prone to cyber-attacks through their sensors and communication links. The injection of false data packets in the cyber layer can disrupt the control objectives, leading to voltage instability and load sharing patterns. Therefore, detection and correction of malicious data is essential for the DC microgrid stability. In this article, a DDNN is designed with prediction and correction networks. The prediction network composed with one input layer, two hidden layers and one output layer. This network predicts the converter's duty by considering the input features as DC bus voltage and the reference voltage. The correction network also composed with one input layer, two hidden layers and one output layer. This network provides the duty corresponding to the attack by considering the input features as DC bus voltage, battery voltage and reference voltage. The output from the prediction and correction network are implanted to detect and correct the false data injection (FDI) attacks. However, for the training purpose, the data is collected by performing the various attack scenarios who is able to inject the false data and disrupt the stable operation of the system. The data is then used to train a neural network to detect a larger set of FDI attacks. The proposed scheme's effectiveness is verified by conducting the real-time experiments for various attack scenarios and its results are explored.