A Data-Driven Cyber-Resilience Framework With Minimal Feature Learning for Frequency Restoration in Isolated Networked Microgrids

Isolated networked microgrids (INMGs) provide a sustainable solution for future energy demands by leveraging modern communication technologies to facilitate efficient coordination among isolated microgrids (IMGs). However, reliance on communication networks brings cyber vulnerabilities, potentially undermining the overall performance and reliability of INMGs. Existing cyber-resilience solutions for IMGs are often inflexible, computationally expensive, and lack the adaptability needed for cyber-attack detection and mitigation. These challenges make them unsuitable for INMGs, where the interconnected IMG system complicates the detection of attacks. This paper introduces a scalable, data-driven intelligent framework designed for the concurrent detection and mitigation of attacks targeting the upper control layers of the INMG. The framework employs sparse Bayesian learning via the sequential Monte Carlo method to estimate the posterior distribution weights of the dynamic neural network model. This approach enhances sparsity by prioritizing key weights and filtering out insignificant ones, resulting in faster error estimation. This error identifies attacks, compromised IMGs, and communication channels, triggering timely attack mitigation for frequency restoration and cost-efficient operation of the INMG. Compared to existing strategies, the proposed approach offers significant improvements in computational efficiency and effectively manages the complexity of simultaneous attacks on multiple communication channels, distinguishing between malicious interference and normal system fluctuations, and enabling faster real-time mitigation. The proposed framework is validated using an INMG frequency-control model simulated in MATLAB and OPAL-RT under various attack scenarios. Results demonstrate its effectiveness in managing cyber-attacks and enhancing the adaptive frequency resilience of the INMG.