基于订单扰动的炼钢-连铸区段Q-learning遗传优化重调度模型  

Q-learning-based genetic optimization rescheduling model for steelmaking-continuous casting section with order disturbance

作　　者：陈虹志 邵鑫 刘青 张江山[1] 高山 CHEN Hongzhi;SHAO Xin;LIU Qing;ZHANG Jiangshan;GAO Shan(State Key Laboratory of Advanced Metallurgy,University of Science and Technology Beijing,Beijing 100083,China;Engineering Research Center of Manufacturing Executing System Technology for Iron and Steel Production,University of Science and Technology Beijing,Beijing 100083,China;Laiwu Iron and Steel Group Yinshan Section Steel Co.,Ltd.,Jinan 271104,Shandong,China)

机构地区：[1]北京科技大学钢铁冶金新技术国家重点实验室,北京100083 [2]北京科技大学钢铁生产制造执行系统教育部工程研究中心,北京100083 [3]莱芜钢铁集团银山型钢有限公司,山东济南271104

出　　处：《钢铁》2023年第11期90-99,共10页Iron and Steel

基　　金：国家自然科学基金资助项目(52374321)。

摘　　要：当今市场对钢铁产品的需求呈现多品种、小批量、多规格和高质量的特点,导致炼钢厂订单复杂,采用异钢种连浇技术能够有效提高连铸连浇炉次,减少完成合同所需要的浇次数量,提高生产连续性。受生产订单的驱动,炼钢-连铸区段常会因紧急订单的出现而干扰原有生产计划。根据紧急订单的钢种以及所需要的生产工艺与钢厂的产能、生产工艺、产品库存和原料库存等实际生产情况的匹配程度制定了应对策略,建立了以生产计划内最小总等待时间、最小总延期时间以及最小完工时间为优化目标的炼钢-连铸区段重调度模型,提出了基于Q-learning的改进遗传算法进行求解。采用某炼钢厂2种典型生产模式,即异钢种连浇层流生产模式和四炉对三机同钢种紊流生产模式进行模拟试验,结果表明所建模型能够有效解决订单扰动重调度问题,减少因紧急订单带来的生产计划内总等待时间的增加,减少生产计划的延期时间。通过求解最优排序问题来检验基于Q-learning改进的遗传算法的性能,相比经典遗传算法,基于Q-learning改进的遗传算法能够找到更符合优化目标的最优解,获得最优解的迭代次数更少,相比自适应遗传算法,基于Q-learning改进的遗传算法运行时间减少95.37%,以上结果表明基于Q-learning改进的遗传算法具有良好的求解性能。The demand for steel products in today′s market is characterized by multiple varieties,small batches,multiple specifications,and high quality,resulting in complex orders for steelmaking plants.The use of different steel grade continuous casting technology can effectively increase the number of continuous casting furnaces,reduce the number of castings required to complete contracts,and improve production continuity.Driven by production orders,the steelmaking-continuous casting section often interferes with the original production plan due to the emergence of emergency orders.This article develops a response strategy based on the matching degree between the steel grades of emergency orders and the required production process with the actual production situation of one steelmaking plants,such as production capacity,production process,product inventory,and raw material inventory.A rescheduling model for steelmaking-continuous casting section is established with the optimization objectives of minimum total waiting time,minimum total delaying time,and minimum completion time in the production plan.An improved genetic algorithm based on Q-learning is proposed for solution.This study conducted simulation experiments using two typical production modes in a certain steelmaking plant,which are different steel grade sequence casting laminar production mode and 4BOF-3CCM turbulent production mode producing the same steel grade.The results showed that the established model can effectively solve the problem of orders disturbance rescheduling,reduce the increase in total waiting time in the production plan caused by emergency orders,and reduce the delaying time of the production plan.This article tests the performance of Q-learning based genetic algorithms by solving the optimal sorting problem.Compared to classical genetic algorithms,Q-learning based genetic algorithms can find the optimal solution that is more in line with the optimization objective,with fewer iterations to obtain the optimal solution.Compared to self-adapti

关 键 词：炼钢-连铸 生产调度 订单扰动 Q-LEARNING 遗传算法 异钢种连浇 

分 类 号：TF777[冶金工程—钢铁冶金] TP18[自动化与计算机技术—控制理论与控制工程]