Intelligent Intrusion Detection Scheme for Smart Power-Grid Using Optimized Ensemble Learning on Selected Features

The smart grid has gained a reputation as the advanced paradigm of the power grid. It is a complicated cyber-physical system that combines information and communication technology (ICT) with a traditional grid that can remotely control operations. It provides the medium for exchanging real-time data between the company and users through the advanced metering infrastructure (AMI) and smart meters. However, smart grids have many security and privacy concerns, such as intruding sensitive data, firmware hijacking, and modifying data due to the high reliance on ICT. To protect the power-grid system from these counteracts and for reliable and efficient power distribution, early and accurate identification of these issues needs to be addressed. The intrusion detection in a smart grid system plays an essential role in providing a secure service and transmitting the high priority alert message to the system admin about the detection of adversary attacks. This paper proposes an intelligent intrusion detection scheme to accurately classify various attacks on smart power grid systems. The proposed scheme used the binary grey wolf optimization-based feature selection. It optimized the ensemble classification approach to learn the non-linear, overlapping, and complex electrical grid features taken from publicly available Mississippi State University and Oak Ridge National Laboratory (MSU-ORNL) dataset. The experimental results using a 10-fold cross-validation setup and selected feature subset for two class and three class problems reveal the proposed method's promising performance. Further, the significantly superior performance compared to the existing benchmark methods justified the robustness of the proposed scheme.