Enhanced False Data Injection Attack Detection using Performance-Aware Deep Learning Model for Security Analysis

Abstract

One of the general attacks in cyber-physical systems, the false data injection attack (FDIA), is considered a major complex threat for state estimation. Unlike other attacks, namely distributed denial of service (DDoS) and jamming, effective FDIA can evade the traditional residual- driven poor data detection frameworks. To initiate the FDIA, the attacker uses the vulnerabilities of the communication system at the cyber level and then fine-tunes the measurement of physical devices. Recently, various forms of FDIA have been established, and the attacker can utilize network vulnerabilities. Nowadays, deep learning (DL)-based models are applied to detect FDIA that accept the spatial data features in the system state of a one-time sample to recognize attacks. In this manuscript, we propose an Enhanced False Data Injection Attack Detection with Mutual Information and Stacked Autoencoder (EFDIAD-MISAE) model. Initially, min-max scaling is applied for data pre-processing. In addition, mutual information is leveraged for feature selection. For the attack classification process, the stacked autoencoder method is used. Ultimately, the coati optimization algorithm is applied for tuning. The results of the EFDIAD-MISAE model were experimentally compared with the FDIA data, achieving an accuracy of 99.15%. Literature-based benchmarks were derived from comparative results reported in previous studies, as the experimental datasets and procedures may differ from those of the present study.