Joint optimization of steel rolling process and power demand response: a deep reinforcement learning approach

This paper proposes a novel joint optimization framework for the steel rolling process and power demand response, utilizing a Deep Q-Learning Network (DQN) to solve a nonlinear, bi-objective scheduling problem with multiple operational constraints and dynamic electricity pricing considerations. This method comprises two closely coupled optimisation layers. The first layer minimises the production penalty by intelligently sequencing slabs within the rolling unit, while respecting production constraints. The second layer adjusts rolling schedules according to time-of-use electricity rates to reduce power costs. In doing so, it balances production needs with load management. Case studies show that, compared with genetic algorithms,the proposed method reduces electricity costs by approximately 53.3% and carbon emissions by approximately 47.8%. It also increases renewable energy utilisation by about 3.8%. These results confirm the method’s effectiveness in unlocking rolling mill load flexibility, whilst balancing economic performance with low-carbon objectives.