A Reinforcement Learning-Based Framework for Tariff-Aware Load Shifting in Energy-Intensive Manufacturing.

Area of Science:

• Industrial Engineering
• Artificial Intelligence
• Energy Management

Background:

• Energy-intensive manufacturing faces challenges optimizing costs under time-varying electricity tariffs.
• Existing scheduling strategies often struggle to balance economic benefits with operational feasibility.

Purpose of the Study:

• To develop and validate a tariff-aware load-shifting framework for industrial energy management.
• To leverage industrial sensing data and reinforcement learning for cost reduction.

Main Methods:

• A Proximal Policy Optimization (PPO) reinforcement learning agent was trained in a custom Gymnasium environment.
• The framework utilized time-series measurements of active power and energy from a quicklime plant.
• Load-shifting actions were constrained to 80-125% of a baseline, targeting peak TOU windows.

Main Results:

• A median total cost reduction of approximately 10% was achieved over 30 working days.
• Reductions were driven by decreased energy consumption and demand peaks during critical hours.
• Deviations from energy balance and minimum production targets were observed, indicating a cost-production trade-off.

Conclusions:

• Reinforcement learning, specifically PPO, offers a competitive approach for tariff-aware industrial energy scheduling.
• The framework demonstrates the potential of integrating learning-based decision-making with industrial sensing.
• Further refinement of constraint handling is necessary for robust industrial application.