Improving efficiency of steel converter facility: a digital twin and reinforcement learning approach
提出结合仿真建模、数字孪生和深度强化学习的方法，优化转炉车间天车调度，减少瓶颈等待时间，平均结果接近最优的3.6%，适用于钢铁及其他复杂调度行业。
The steel converter facility is an essential part of the steel manufacturing process and is responsible for converting molten iron into steel. The material movement within the facility depends on the effective overhead crane scheduling. However, the traditional approach of crane scheduling at these facilities leads to inefficiencies. To understand this need, we propose an integrated approach using Simulation modelling, Digital Twin (DT) and Reinforcement Learning (RL) to improve the crane scheduling operation of the steelmaking process. We developed a simulation model using discrete event and agent-based simulation and connected it with real-time data by applying the digital twin approach. Further, a Deep Reinforcement Learning (DRL) agent is trained in a stochastic environment with multiple feedback loops and complex interactions for crane scheduling. The DRL agent recommends the sequences of crane movement to reduce the wait time in bottleneck operation. The experimental results demonstrate that the proposed method gives results within 3.6% of the optimal on average. A case of the steel converter facility illustrates this approach, which is also applicable to other industries requiring complex scheduling, such as construction, automotive, and logistics, to improve operational efficiency.