【作者】 程锦君； 邓南阳； 李家乐； 韦文钰； 耿毅； 张立强；

【Author】 CHENG Jinjun;DENG Nanyang;LI Jiale;WEI Wenyu;GENG Yi;ZHANG Liqiang;School of Metallurgical Engineering, Anhui University of Technology;Long Products Division, Ma’anshan Iron & Steel Co., Ltd.;

【机构】 安徽工业大学冶金工程学院； 马鞍山钢铁股份有限公司长材事业部；

【摘要】 在马钢连铸生产中,重量波动大及模糊的坯料交接界面计量,加之智能化程度不足,均是导致重型异形坯成材率低下的显著因素。针对这一问题,本研究通过采纳工业大数据的全量化分析方法,运用先进的冶金数据库在线优化学习、深度数据挖掘技术及人工神经网络算法,成功构建了重型异形连铸钢坯的定重在线模型。该模型采用分层式架构设计,包含数据采集层、模型计算层、决策控制层和执行反馈层四个核心层次,通过多传感器实时采集中包温度、拉速、结晶器液面、二冷水流量等关键工艺参数,建立了完整的冶金数据库,并开发了切割长度动态修正算法、基于红外摄像的定尺修正算法以及改进的RBF神经网络预测算法,实现了铸坯定重的多传感器融合动态修正。改进的RBF算法采用自适应聚类和动态权重调整策略,通过自适应K-均值聚类方法自动确定最优的隐藏层神经元数量和中心位置,相比传统方法预测误差降低了35%以上,收敛速度提升了50%以上。系统构建了基于Hadoop生态系统的冶金大数据平台,通过深度特征提取与工程优化和多传感器数据融合的动态修正模型,采用改进的卡尔曼滤波算法实现多源异构数据的优化融合,通过精确控制定尺切割显著提升了定重精度,并开发了动态模型优化算法以计算铸坯的确切长度与重量,实现了生产流程的实时预测与闭环控制。该智能定重模型于2023年6月在马钢投入运行后,结果表明重型异形钢的成材率得到了显著提升,铸坯重量公差小于千分之三的最高合格率达到了93%以上,平均合格率稳定在82%以上,而在千分之五范围内的合格率达到了99%以上。该模型以其动态调整、智能修正、自我学习优化及实时预判的特性,有效改善了因铸坯重量造成的轧材短尺或切废现象,在节能降耗和提高产品成材率方面发挥了显著作用,为重型异形坯连铸生产的智能化升级提供了重要技术支撑。更多还原

【Abstract】 In the continuous casting production of Masteel, significant weight fluctuations and ambiguous billet handover interface measurement, coupled with insufficient intelligent automation, are prominent factors contributing to the low yield rate of heavy special-shaped billets. To address this issue, this study adopts a comprehensive analysis approach based on industrial big data, utilizing advanced metallurgical database online optimization learning, deep data mining technology, and artificial neural network algorithms to successfully construct an online weight-fixing model for heavy special-shaped continuous casting billets. The model employs a hierarchical architecture design comprising four core layers: data acquisition layer, model calculation layer, decision control layer, and execution feedback layer. Through multi-sensor real-time acquisition of key process parameters including tundish temperature, casting speed, mold level, secondary cooling water flow, and other critical operational parameters, a comprehensive metallurgical database is established. The system develops dynamic cutting length correction algorithms, infrared camera-based fixed-length correction algorithms, and improved RBF neural network prediction algorithms to achieve multi-sensor fusion dynamic correction for billet weight fixing. The improved RBF algorithm adopts adaptive clustering and dynamic weight adjustment strategies, automatically determining the optimal number of hidden layer neurons and center positions through adaptive K-means clustering methods,achieving over 35% reduction in prediction error and 50% improvement in convergence speed compared to conventional methods. The system constructs a metallurgical big data platform based on the Hadoop ecosystem, implementing deep feature extraction and engineering optimization along with multi-sensor data fusion dynamic correction models, utilizing improved Kalman filtering algorithms to achieve optimal fusion of multi-source heterogeneous data. Through precise control of fixed-length cutting, the weight-fixing accuracy is significantly enhanced, and dynamic model optimization algorithms are developed to calculate the exact length and weight of cast billets, achieving real-time prediction and closed-loop control of the production process. After the intelligent weight-fixing model was implemented at Masteel in June 2023, results demonstrate that the yield rate of heavy special-shaped steel has been significantly improved, with the highest qualification rate reaching over 93% for billet weight tolerance within three parts per thousand, average qualification rate stabilized above 82%, and qualification rate within five parts per thousand reaching over 99%. The model, with its characteristics of dynamic adjustment, intelligent correction, self-learning optimization, and real-time prediction, effectively ameliorates the phenomena of rolled material short lengths or cutting waste caused by billet weight variations, playing a significant role in energy conservation, consumption reduction, and improving product yield rate, providing crucial technical support for the intelligent upgrading of heavy special-shaped billet continuous casting production.更多还原