Condition-triggered adaptive robust MPC for flexible load regulation in energy-intensive industries: A case study of aluminum electrolysis

Energy-intensive industries face a growing conflict between power-system flexibility requirements and the stringent stability constraints of continuous electro-thermal processes. To overcome the limited adaptability of static scheduling and controllers with fixed conservatism under multi-regime operating conditions, this study proposes a condition-triggered robust model predictive control (CT-RMPC) framework. The method couples a multi-energy hub representation with distribution-free conformal prediction to calibrate uncertainty bounds and introduces a trigger layer that adjusts the robust budget Γ online according to regime indicators, thereby trading off economic performance and risk-aware feasibility. Industrial validation is conducted on a 300kA aluminum electrolysis line using 18 months of high-resolution operational data. The proposed controller reduces the peak grid import power from 160 MW to 151 MW (−5.6%) while maintaining the specific energy consumption within 13.2 ± 0.2 MWh∙t−1 Al. The validated period yielded a cumulative cost savings of 652,000 USD, and the monthly unit net effect remains positive, ranging from 0.4 to 3.8 USD·t−1Al, with demand-charge reduction contributing more than 35% of the total benefit. Although validated on aluminum electrolysis, the framework is formulated in a modular manner and can be extended to other EIIs with strong thermal inertia and multi-energy coupling, provided that process-specific states and feasibility constraints are re-instantiated and validated.