AI-enabled cybersecurity framework for three-phase grid-tied inverters

Abstract

Grid-tied inverters, which connect renewable energy sources such as solar and wind installations to the utility grid, are increasingly exposed to cyber threats as power systems become more digitalised and remotely accessible. Unlike conventional information technology networks, these devices operate under strict real-time control requirements and physical safety constraints, which makes adapting general-purpose cybersecurity tools non-trivial. This paper proposes an artificial intelligence based intrusion detection framework specifically designed for three-phase grid-tied inverter systems. The framework addresses four distinct attack categories: false data injection targeting sensor measurements, command injection manipulating reactive power control, denial-of-service disrupting communication channels, and coordinated attacks that degrade multiple system parameters simultaneously. A set of electrical and operational features is extracted from time-windowed inverter measurements, and five machine learning classifiers are combined through a weighted ensemble strategy to improve detection reliability over any individual model. The ensemble achieves competitive detection accuracy and sub-second response times suitable for protective relay coordination, while maintaining a low false alarm rate that is important for operator confidence in practical deployment. The feature selection methodology, window parameter sensitivity, and ensemble weight optimisation are each examined to provide transparent justification for design choices. Limitations of the current work are explicitly acknowledged, including the exclusive use of software simulation, the absence of hardware-in-the-loop validation, and the focus on non-covert attack variants. These limitations define a clear agenda for future experimental work. The study contributes a structured and reproducible methodology for evaluating machine learning based cybersecurity frameworks in power electronics, a domain where such systematic treatments remain scarce.