东北大学学报:自然科学版 ›› 2018, Vol. 39 ›› Issue (12): 1717-1722.DOI: 10.12068/j.issn.1005-3026.2018.12.009

• 材料与冶金 • 上一篇    下一篇

遗传算法优化神经网络的热轧带钢弯辊力预报模型

王振华, 龚殿尧, 李广焘, 张殿华   

  1. (东北大学 轧制技术及连轧自动化国家重点实验室, 辽宁 沈阳110819)
  • 收稿日期:2017-08-21 修回日期:2017-08-21 出版日期:2018-12-15 发布日期:2018-12-19
  • 通讯作者: 王振华
  • 作者简介:王振华(1990-),男,山西盂县人,东北大学博士研究生; 张殿华(1963-),男,内蒙古赤峰人,东北大学教授,博士生导师.冯明杰(1971-), 男, 河南禹州人, 东北大学副教授; 王恩刚(1962-), 男, 辽宁沈阳人, 东北大学教授,博士生导师.
  • 基金资助:
    国家自然科学基金资助项目(51171041).国家重点研发计划项目(2017YFB0304100); 国家自然科学基金资助项目(51634002, 51774084, 51704067).

Bending Force Prediction Model in Hot Strip Rolling Based on Artificial Neural Network Optimize by Genetic Algorithm

WANG Zhen-hua, GONG Dian-yao, LI Guang-tao, ZHANG Dian-hua   

  1. State Key Laboratory of Rolling and Automation, Northeastern University, Shenyang 110819, China.
  • Received:2017-08-21 Revised:2017-08-21 Online:2018-12-15 Published:2018-12-19
  • Contact: WANG Zhen-hua
  • About author:-
  • Supported by:
    -

摘要: 针对传统弯辊力预设定模型的缺陷和带钢热连轧轧制特点,利用某钢铁公司1580mm热轧线生产数据,对精轧机组末机架进行了基于遗传算法优化神经网络的弯辊力预报模型研究.以大量实际数据作为神经网络训练输入,充分考虑了输入参数之间的影响作用,模型结构简单、容易实现,其整体性能用平均绝对百分误差、均方根误差和相关系数R评价.通过将预测结果与实测结果比较,验证了模型的精度.研究发现,提出的弯辊力预测模型相比于传统模型可实现高度非线性拟合,适用于提高热轧带钢头部板形控制精度,为实际弯辊力设定提供指导和试验基础.

关键词: 神经网络, 热轧板形, 生产数据, 弯辊力, 遗传算法

Abstract: In view of the drawbacks of the traditional bending force preset model and the characteristics of hot strip rolling, a bending force prediction model is built based on genetic algorithm(GA)and artificial neural network(ANN)and the prediction is carried out on the end stand of finishing mill using a large amount of production data from 1580mm hot rolling line in a steel corp. The features of the model are as follows: a large number of actual data is used as input for the ANN training, the influence of various input parameters are fully considered, the framework of the model is relatively simple and easy to implement, and the overall performance is evaluated by the mean absolute percentage error, root mean square error and correlation coefficient. By comparing the predicted results with the experimental ones, the prediction accuracy of the model is verified. It shows that the GA-ANN prediction model of roll bending force can realize a high nonlinear fitting, which is suitable for improving the accuracy to control head shape of hot strip rolling. This study provides guidance and test foundation for the actual bending force setting.

Key words: artificial neural network, shape of hot strip rolling, production data, bending force, genetic algorithm

中图分类号: