Plant-wide optimization integrating inside- and outside-battery-limit processes: A modular AI framework with self-supervised, active, and hybrid learning

초록

Large-scale manufacturing facilities consist of complex systems integrating primary production units (inside battery limits) and auxiliary utility systems (outside battery limits). To maximize overall plant efficiency and environmental sustainability, a unified optimization approach for these highly interconnected processes is essential. This study proposes a comprehensive AI-driven optimization framework for integrated manufacturing processes based on large-scale, real-world industrial datasets. The framework incorporates three synergistic methodologies: (1) self-supervised learning to optimize operation schedules by identifying latent patterns in historical data; (2) active learning to strategically explore optimal operating conditions and enhance model robustness by targeting regions of high uncertainty; and (3) hybrid modeling, which integrates domain-specific physical principles with data-driven algorithms for precise equipment maintenance and emission forecasting. The framework was validated through on-site industrial application, demonstrating that historical operating cases offer up to 2.97% reduction in relative energy consumption (approximate to 3,182 t CO2-eq/year) when operators select the recommended schedule. The active learning module successfully expanded the searchable operating envelope into previously unexplored regimes, achieving a high predictive accuracy with an R2 exceeding 0.98. Furthermore, the results confirmed the framework's effectiveness in forecasting emissions and managing purification system efficiency with a reliable performance of an R2 of 0.75. This research contributes to process systems engineering by: 1) streamlining operation schedules, 2) effectively broadening feasible operating conditions, and 3) bridging the gap between air pollutant management and equipment efficiency. Ultimately, this framework provides a reliable and scalable solution for integrated process optimization, advancing the development of sustainable and efficient large-scale manufacturing operations.

키워드