A New Accelerated Attentive Deep Learning-Based Approach to Early Detect Attacks in Cyber-Physical Microgrids

Abstract

The integration of renewable energy resources (RERs) and electric vehicles (EVs) into microgrids enables the provision of ancillary services for frequency and voltage regulation, thus improving the stability and efficiency. However, such integration requires a set of communication networks to exchange information among the microgrid components, which makes the microgrid assets prone to cyber vulnerability threats. Unlike in previous work, in which existing approaches wait until the impacts appear on the system to be able to detect the attacks, this paper introduces a new approach that combines the opening image technique and attentive deep learning to early detect the cyberattacks applied to a cyber-physical microgrid embedded with RERs and EVs. Furthermore, this paper investigated the effect of smart meters’ data granularity on the attack detection accuracy. The results have shown that the use of a high time resolution of 1-sec increases the detection accuracy reaching 99.91%. The training process of the proposed approach has been accelerated using Graphics Processing Unit (GPU), which demonstrated low computational time by significantly reducing both the training and testing time by 93% and 70% respectively.