Manufacturing plants are major energy consumers with complex and dynamically changing energy flows. A key element of sustainable manufacturing is real-time, precise energy scheduling to handle volatile electricity prices, intermittent renewable generation, and changing production demands. Traditional energy-management systems lack automation and struggle to perform multi-objective coordinated optimization.

This proposal builds an integrated energy “brain” of sensing → forecasting → optimization → execution. The LLM plays the sensing and forecasting roles: it analyzes unstructured textual inputs such as weather forecasts (which affect shop-floor temperature control), production schedules, and grid demand-response signals to accurately predict factory load and on-site renewable (e.g., rooftop PV) generation over the next hours or days. Then the intelligent optimization (IO) algorithm takes over to make optimization decisions: targeting minimal total energy cost and minimal carbon emissions while meeting production deadlines, it dynamically adjusts shiftable loads (e.g., washing and testing tasks), the on/off timing of high-power equipment, and the charge/discharge strategy of energy-storage systems in real time.

For example, on foreseeing an afternoon price peak and a grid peak-shaving request, the system can automatically postpone electroplating to nighttime and draw power from battery energy storage, thereby lowering electricity costs and participating in grid ancillary services to support system-level stability. This LLM–IO synergy elevates energy management from passive monitoring to active optimization, achieving both economic and environmental benefits and serving as a core pillar of a green transformation for intelligent manufacturing plants.