摘要
针对带钢热镀锌板锌层厚度偏差易受产线多变量强耦合和测厚仪滞后时间长等因素影响的问题,提出一种基于最大信息系数(MIC)以及改进的粒子群算法优化正则化极限学习机(IPSO-RELM)的带钢锌层厚度预测方法。首先,采集生产过程数据进行相关预处理;然后,利用MIC法对各参数变量进行重要性排序,确定影响锌层厚度的关键因素;最后,将筛选的变量作为输入项建立RELM预测模型,并通过IPSO算法优化模型的随机性参数,有效提高了模型的稳定性和预测精度。结果表明:所建立模型预测结果的拟合决定系数R^(2)为94.66%,预测误差在-4~4 g/m^(2)的样本点命中率达到96%,且模型的3项评价指标均优于其他对比算法,证明了所提方法预测精度高,可为带钢热镀锌板产品质量的提升奠定基础。
Aiming at the problem that the thickness deviation of zinc layer of hot-dip galvanized strip sheet is affected by factors such as multi-variable strong coupling of production line and long lag time of thickness gauge,a thickness prediction method of zinc strip layer based on maximum information coefficient(MIC)and improved particle swarm algorithm optimization regularized extreme learning machine(IPSO-RELM)was proposed.Firstly,the production process data was collected for relevant preprocessing.Then,the MIC method was used to rank the importance of each parameter variable to determine the key factors affecting the thickness of the zinc layer.Finally,the RELM prediction model was established by taking the filtered variables as inputs,and the randomness parameters of the model were optimized by the IPSO algorithm,which effectively improved the stability and prediction accuracy of the model.The results show that the fitting coefficient of determination R2 of the predicted results of the established model is 94.66%,the hit rate of the sample points with the prediction error of-4-4 g/m^(2) reaches 96%,and the three evaluation indicators of the model are better than other comparison algorithms,which proves that the proposed method has high prediction accuracy and can lay a foundation for the improvement of the quality of strip hot-dip galvanized sheet.
作者
方军
王兴东
汪洋
吴宗武
丁健
FANG Jun;WANG Xingdong;WANG Yang;WU Zongwu;DING Jian(Key Laboratory of Metallurgical Equipment and Control Technology,Wuhan University of Science and Technology,Wuhan 430081,China;Baoxin Software Wuhan Co.,Ltd.,Wuhan 430080,China;Baosteel Wuhan Iron and Steel Co.,Ltd.,Wuhan 430083,China)
出处
《热加工工艺》
北大核心
2024年第22期62-68,82,共8页
Hot Working Technology
基金
国家自然科学基金资助项目(51875418)。
关键词
带钢热镀锌
锌层厚度预测
最大信息系数
改进的粒子群算法
正则化极限学习机
hot-dip galvanized strip
zinc layer thickness prediction
maximum information factor
improved particle swarm arithmetic
regularized extreme learning machine
