Real-time Optimization for Wind-to-H2 Driven Critical Infrastructures: High-fidelity Active Constraints and Integer Variables Prediction Enhanced by Feature Space Expansion

This paper focuses on developing a real-time optimal operation model for a new engineering system, wind-to-hydrogen-driven low-carbon critical infrastructure (W2H-LCCI), that utilizes wind power to generate hydrogen through electrolysis and combines it with carbon capture to reduce carbon emissions from the power sector. First, a convex mathematical model for W2H-LCCI is proposed, and then optimization models for its real-time decision-making are developed, which are mixed-integer convex programs (MICPs). Furthermore, since this large-scale MICP problem must be solved in real-time, a fast solution method based on active constraint and integer variable prediction (ACIVP) is presented. ACIVP method predicts the binary variable values and the set of limited-number constraints, which most likely contain all of the active constraints, based on historical optimization data. It results in only a small-scale continuous convex optimization problem needing to be solved by optimization solvers for W2H-LCCI real-time optimal operation. To increase the accuracy of the ACIVP method, feature space expansion (FSE) is employed, and a multi-stage ACIVP-FSE method is proposed. The effects of stage design and stage ordering on ACIVP-FSE performance are also discussed. We validate the effectiveness of the developed system and solution method using two water-energy nexus case studies.