Answers are grounded in this paper and cite their sources.
Heterogeneous industrial clusters such as refiners, data centers, and manufacturing facilities represent high-density loads that should maintain operational continuity during grid emergencies while providing flexible frequency recovery support. Existing strategies-conventional under-frequency load shedding, centralized, and independent controllers- either result in disproportionate energy curtailment across clusters, imposing heavy communication overhead, or are reactive in nature. This paper introduces a fault-triggered vulnerability-aware distributed Model Predictive Control (MPC) architecture in which each cluster autonomously optimizes its operation during normal conditions, and a consensus layer based on the Alternating Direction Method of Multipliers (ADMM) activates only when frequency deviates from a defined threshold. A hysteresis approach prevents spurious activation and deactivation around the threshold. The introduced vulnerability-aware mechanism results in an adaptive consensus that redistributes the burden away from the clusters approaching cascading failure limits. Voltage constraints are enforced via linearized sensitivities derived from a Newton-Raphson AC power flow. Validation across a 24-hour simulation with three heterogeneous faults, four comparison baselines, five uncertainty scenarios, and cluster counts up to 10, demonstrated that the proposed framework reduces energy curtailment by 68% and 56% relative to independent and centralized MPC, respectively. The Jain fairness index of 0.983 confirms that the proposed architecture results in optimal burden distribution across heterogeneous clusters. The ADMM coordination operates with a duty cycle of 11.2% only, and the convergence rate improves from 72% to 85% from two to ten clusters, illustrating scalability. All frequency deviations remain within operational bounds well above the under frequency relay thresholds defined in IEEE Std C37.117-2007.
No takes yet. Share an insight, caveat, or question.
Shaukat et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: