Machine Learning-Accelerated Method for Real-Time Optimization of Micro Energy-Water-Hydrogen Nexus

This paper explores the micro Energy-Water- Hydrogen (m-EWH) nexus, an engineering system designed to reduce carbon emissions in the power sector. The m-EWH nexus leverages renewable energy sources (RES) to produce hydrogen via electrolysis, which is then combined with carbon captured from fossil fuel power plants to mitigate emissions. To address the uncertainty challenges posed by RES, this paper proposes a real-time decision-making framework for the m-EWH nexus, which requires the rapid solution of large-scale mixed-integer convex programming (MICP) problems. To this end, we develop a machine learning-accelerated solution method for real-time optimization (MARO), comprising three key modules: (1) an active constraint and integer variable prediction module that rapidly solves MICP problems using historical optimization data; (2) an optimal strategy selection module based on feasibility ranking to ensure solution feasibility; and (3) a feature space extension and refinement module to improve solution accuracy by generating new features and refining existing ones. The effectiveness of the MARO method is validated through two case studies of the m-EWH nexus, demonstrating its capability to swiftly and accurately solve MICP problems for this complex system.