基于非线性自回归神经网络的烧结混合料水分智能控制  

Intelligent control of sintering mixture moisture based on NARX neural network

作　　者：吴岳明[1] 聂慧远[2] 吴朝霞 WU Yueming;NIE Huiyuan;WU Zhaoxia(Equipment Department,Shanghai Meishan Iron and Steel Co.,Ltd.of Baosteel,Nanjing 210039,China;Ironmaking Plant,Shanghai Meishan Iron and Steel Co.,Ltd.of Baosteel,Nanjing 210039,China;School of Control Engineering,Northeastern University at Qinhuangdao,Qinhuangdao 066004,China)

机构地区：[1]宝钢股份上海梅山钢铁股份有限公司设备部,江苏南京210039 [2]宝钢股份上海梅山钢铁股份有限公司炼铁厂,江苏南京210039 [3]东北大学秦皇岛分校控制工程学院,河北秦皇岛066004

出　　处：《冶金自动化》2021年第1期27-33,共7页Metallurgical Industry Automation

基　　金：河北省自然科学基金资助项目(F2017501041);中央高校基本科研业务费资助项目(N172304030)。

摘　　要：为提高烧结混合料加水控制的智能程度,针对烧结生产线现场的实际情况,提出了一种非线性自回归(nonlinear autoregressive models with exogenous inputs,简称NARX)神经网络加水前馈控制模型。首先,利用工厂的历史生产数据和物料平衡原理建立烧结混合料的加水模型,其中涉及各种原料量及混合加水量;然后,将深度监督学习与自学习NARX算法结合,建立烧结混合料水分预测模型;最后,将NARX水分预测模型与极限学习机(extreme learning machine,简称ELM)模型进行对比。结果表明,NARX算法在水分控制的准确性、平稳性上均更优,可以为烧结混合料加水前馈控制提供新的解决方案。To improve the intelligence of the moisture control of the sintering mixture,a NARX neural network water feedforward control model was proposed for the actual situation of the sintering line.Firstly,the plant’s historical production data and material balance principle are used to establish the sintering mixture adding water model,which involves various raw material quantities and mixed water addition. Then,the depth supervision learning is combined with the self-learning NARX algorithm to establish the moisture prediction model of the sintering mixture. Finally,compared with the ELM model,the results show that the NARX algorithm is superior in the accuracy and stability of moisture control,and can provide a new solution for water feedforward control of sintering mixture.

关 键 词：烧结混合料 深度学习 非线性自回归 前馈控制 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TF046.4[自动化与计算机技术—控制科学与工程]