Enhancing Cybersecurity and Resilience in Distribution Networks via Hybrid Partitioning and GAN-Driven Dynamic Reconfiguration

This paper presents a structured hybrid grid partitioning framework designed to enhance cyber-physical resilience and scalability in distribution networks, particularly under high electric vehicle (EV) penetration and evolving cyber threats. The framework integrates three tightly coupled layers. First, a graph-based clustering stage introduces spectral-informed adaptive hierarchical clustering (SIAHC), which combines global spectral features with composite electrical distance metrics to generate modular, self-sufficient, and topologically coherent subnetworks, supporting power loss minimisation, voltage stability, and topological robustness. Second, an optimisation-based refinement layer employs the alternating direction method of multipliers (ADMM) for scalable, distributed coordination across partitions, ensuring feasibility under power flow and voltage constraints. Third, a feedback-informed data-driven layer integrates a Bayesian LSTM variational autoencoder (LSTM-VAE) to forecast cost components and learn clustering parameters, and a conditional Wasserstein GAN (cWGAN-GP) to simulate adversarial scenarios and enable adaptive fallback control under cyber intrusions. The framework is validated on IEEE 33-bus and 123-bus systems with high penetration of EVs, energy storage, and photovoltaics. Results demonstrate improved clustering quality, reduced power loss and voltage deviations, fast convergence, and enhanced cyber-resilience.