机器学习在连铸过程应用研究进展
Research progress on application of machine learning in continuous casting process
周乐君 1王万林 1计熠 1陈佳希1
作者信息
- 1. 中南大学冶金与环境学院,湖南长沙 410083
- 折叠
摘要
在连铸生产中,各工艺参数与控制目标之间存在复杂的非线性关系,通过数值模拟、实验室试验等方法进行优化效率太低,难以满足企业高效生产需求.目前,机器学习已广泛应用于异常预测、质量检测和工艺优化等过程,旨在提高生产效率和铸坯质量,加速连铸新技术的开发和数字化转型.总结梳理了连铸不同生产环节所采取的机器学习预测思路、特征选择和模型构建应用现状,结果表明,与传统方法相比,机器学习在连铸生产中具有更高的预测精度和良好的泛化能力.针对不同预测目标建立相应模型,可以实现连铸过程的精细化和智能化控制.同时,从样本分布、数据质量、模型开发与应用3个方面,对未来机器学习应用于连铸生产的研究方向提出展望,以期为连铸生产智能化发展提供参考.
Abstract
There are complex non-linear relationships between various process parameters and control targets in con-tinuous casting process.Traditional methods,such as numerical simulation and laboratory experiment,cannot meet the requirement of efficient production of enterprises due to their poor optimization efficiency.At present,machine learning has been widely used in the abnormal prediction,quality detection and process optimization with the aims to improve productivity and slab quality,as well as accelerate the development of new technologies and digital transfor-mation in continuous casting.Current advancements of machine learning application in the predicted strategy,fea-ture extraction and model construction in various stages of continuous casting process are summarized.The results show that the machine learning offers higher predictive accuracy and better generalization ability in continuous cast-ing production compared with the traditional methods.It also can achieve fine and intelligent control of continuous casting process after building corresponding models for different prediction targets.Meanwhile,feature research di-rections for machine learning application in continuous casting production are also proposed from three aspects of sample distribution,data quality,and model development/application.This outlook aims to provide valuable refer-ences for the intelligent advancement of continuous casting.
关键词
连铸/机器学习/异常预报/质量检测/工艺优化/智能化Key words
continuous casting/machine learning/anomaly prediction/quality detection/process optimization/in-telligentization引用本文复制引用
出版年
2024
NETL
NSTL
万方数据