Stacking-based multi-objective approach for detection of smart power grid attacks using evolutionary ensemble learning

Smart power grid (SPG) has gained a reputation as the advanced paradigm of the power grid. It provides a medium for exchanging real-time data between the company and users through the advanced metering infrastructure delivering transparent and resilient service to electricity consumers. The widespread deployment of remotely accessible networked equipment for grid monitoring and control has vastly increased the surface of SPG for attackers to locate vulnerable points. The early and accurate identification of the above counteracts is paramount to ensure stable and efficient power distribution. This paper proposes a stacking-based multi-objective evolutionary ensemble scheme to identify various attacks in the SPG. The proposed method used a non-dominated sorting genetic algorithm to learn the non-linear, overlapping, and complex electrical grid features to predict the type of malicious attacks. The experimental results and comparison using multiclass dataset validate the presented 'Stacking-NSGA-II' approach notably outperformed the others benchmark classifiers.