摘要
为准确反映精轧轧制参数和微观组织转变对终轧温度的影响,利用自适应线性神经网络(Adaline)和径向基神经网络(RBF)技术建立了热焓修正系数预报网络作为终轧温度长继承计算模型。首先基于热焓形式的导热偏微分方程建立了带钢终轧温度计算模型,并对带钢在辊缝变形区产生的变形功、摩擦功、与工作辊的接触导热以及机架间冷却换热进行了模型描述;然后从温度与热焓之间的转换关系入手,确定将精轧区域热焓修正系数作为终轧温度模型的自适应参数,并利用复合神经网络技术建立了由19个输入节点,20个RBF隐含层节点,20个Adaline隐含层节点和1个输出节点构成的热焓修正预报网络。结合现场数据,描述了该预报网络训练样本的构成、数据标准化处理方法,同时给出了典型的网络参数和网络的预报能力。
In order to accurately reflect the influence of finishing rolling process parameter and microstructure transformation on finishing rolling temperature,adaptive linear neural network(Adaline)and radial basis neural network(RBF)technology had been utilized to eastablish an enthalpy correction network as long inheritance model of finishing rolling temperature.Firstly,strip temperature computation model was created based on differential equation of heat conduction with enthalpy form.Then deformation work and friction work produced in roll gap deformation zone,thermal conduction between strip and work roll,strip heat transfer with air or spraying water between stands were described with corresponding heat transfer models.Secondly,enthalpy correction coefficient was selected as the adaptive parameter of finishing rolling temperature model according to the conversion relationship between temperature and enthalpy.Then enthalpy correction neural network with 19 input nodes,20 RBF hidden nodes,20 Adaline hidden nodes and one output nodes was created by duplex neural network technology.Based on the field data,the composition of training samples and pattern standardization processing method were revealed.Meanwhile,typical network parameters and predictive ability of the neural network were also presented.
作者
彭良贵
邢俊芳
陈国涛
王小东
张忠伟
PENG Lianggui;XING Junfang;CHEN Guotao;WANG Xiaodong;ZHANG Zhongwei(State Key Laboratory of Rolling Automation,Northeastern University,Shenyang 110819,China;Sheet and Strip Business,Chengsteel Company,HBIS Group,Chengde 067102,China;Automation Center,Chengsteel Company,HBIS Group,Chengde 067102,China)
出处
《轧钢》
2021年第5期75-80,共6页
Steel Rolling
基金
中央高校基本科研业务专项资金项目(N170708020)
辽宁省自然基金资助计划(2020-MS-094)。
关键词
终轧温度
复合神经网络
热焓
自适应模型
传热模型
finishing rolling temperature
duplex neural network
enthalpy
adaptive model
heat transfer model
