ScholarWorks@경상국립대학교

초록

This study presents a machine learning–based optimization framework for utility systems in large-scale manufacturing operations. Designed for broad applicability across diverse industrial processes, the framework integrates historical operational and utility data to support energy-efficient decision-making. Three case studies were conducted to evaluate the effectiveness of the framework. The first case involved identifying feasible operating regions from high-resolution data to optimize utility production in a plant-level utility system. Through this, utility consumption was reduced by 2 %–11 %, resulting in economic efficiency improvements ranging from 6 % to 10 %. The associated reductions in greenhouse gas emissions were also estimated using a life cycle assessment database. The second case applied representation learning techniques to evaluate the optimality of current process operations by comparing them with similar historical instances, offering operational guidance based on data-driven similarity analysis. The third case focused on data storage optimization, where transformation of industrial datasets led to approximately 140-fold reduction in data volume, with implications for integration with image-based AI systems. Together, these case studies demonstrate the potential of machine learning techniques to reduce energy usage, enhance economic performance, and improve data handling in complex manufacturing environments. © 2025 Elsevier Ltd

키워드