基于IHPO-KELM的冷轧带钢板形模式识别  

Pattern Recognition of Cold⁃Rolled Strip Plate Shape Based on IHPO⁃KELM

作　　者：周亚罗 张少川 刘文广[2] 张瑞成 ZHOU Yauo;ZHANG Shaochuan;LIU Wenguang;ZHANG Ruicheng(College of Electrical Engineering,North China University of Science and Technology,Tangshan 063210,Hebei,China;Shougang Jingtang United Iron and Steel Co Ltd,Tangshan 063200,Hebei,China)

机构地区：[1]华北理工大学电气工程学院,河北唐山063210 [2]首钢京唐钢铁联合有限责任公司,河北唐山063200

出　　处：《矿冶工程》2023年第6期162-168,共7页Mining and Metallurgical Engineering

基　　金：河北省自然科学基金(F2018209201);唐山市科技局科技计划项目(22130213G)。

摘　　要：针对目前板形识别方法存在的识别精度低、速度慢等问题,提出了一种改进猎食者算法优化核极限学习机(IHPO-KELM)的冷轧带钢板形识别模型。首先,为减少网络中初始参数的数量、提高板形识别的精度与速度,采用了核极限学习机(KELM)网络;其次,为提高猎食者(HPO)算法的精度,利用基于Sine混沌映射初始化猎食者算法的种群,并针对HPO在迭代过程中易陷入局部早熟的问题,在改进的线性组合位置更新公式中加入莱维飞行机制;然后利用改进猎食者算法对核极限学习机网络识别模型的正则化系数和核参数进行优化,提高板形识别的精度;最后,通过Matlab仿真验证了IHPO-KELM算法具有网络结构简单、收敛速度快、识别精度高等优点。采用IHPO-KELM算法对某公司900HC可逆冷轧机实测数据进行识别,其识别精度比麻雀算法优化KELM(SSA-KELM)识别模型提高了58.8%,表明IHPO-KELM识别模型具有良好的泛化能力,为板形缺陷的高效智能识别提供了新思路。To address the problems of low recognition accuracy and slow speed of current plate shape pattern recognition methods,a pattern recognition model for cold⁃rolled strip plate shape based on IHPO⁃KELM was proposed.Firstly,kernel extreme learning machine(KELM)network was adopted to reduce the number of initial parameters in the network and improve the accuracy and speed of plate shape recognition.Secondly,Levy flight mechanism was added to the improved position update formula for linear combination by using the population of the predator algorithm initialized based on Sine chaos mapping,so as to improve the accuracy of the predator(HPO)algorithm,as well as to address the problem of HPO easily falling into local precocity during the iteration process.Then,the improved predator algorithm was used to optimize the regularization and kernel parameters of the KELM network model,as well as improve the accuracy of plate shape recognition.Finally,the Matlab simulation results have verified that the IHPO⁃KELM algorithm has the advantages of simple network structure,high convergence speed,and high recognition accuracy.The recognition accuracy of IHPO⁃KELM algorithm in identifying the measured data of 900HC reversible cold rolling mill of a company is higher than the KELM recognition model optimized with sparrow algorithm(SSA⁃KELM)by 58.8%,indicating a good generalization ability of IHPO⁃KEM recognition model.This provides a new idea for efficient and intelligent recognition of plate shape defects.

关 键 词：板形缺陷 冷轧带钢 板形识别 改进猎食者算法 神经网络 核极限学习机 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]