Informatics-Centric Neural Network for Distributed Energy Resources Against Diverse Cyber Threats

Abstract

This article addresses challenges in modernizing microgrids (MGs) with distributed energy resources (DERs), which emphasizes cybersecurity vulnerabilities causing from integrating high DERs with cyber-physical data. To ensure seamless integration of DERs and to achieve optimal control performance, this article introduces an informatics-centric neural network (named I-ANN), which is specifically designed for DERs in weak MGs to encounter cyber threats, such as communication latency, false data injection, denial of service, and controller hijacking. The proposed framework utilizes multiagent systems to model the risks posed by cyber threats, with a particular emphasis on their impacts on frequency and voltage regulations. Here, the proposed I-ANN features a novel loss function for automatic signal restoration, and the I-ANN is iteratively trained using various cyber threat scenarios. During critical MG operating scenarios, the new loss function is proposed to enhance robustness and damping while simultaneously mitigating rapid fluctuations in voltage and frequency. Moreover, a significant departure from typical voltage and frequency control loops is the complete replacement of conventional PI controllers with the proposed I-ANN. This strategy fortifies resilience without requiring any additional controllers or control loops. Comparative analyses demonstrate I-ANN's effectiveness in low-inertia MGs with DERs through probabilistic small-signal stability analysis and time-domain simulations under diverse operating conditions.