摘要
为自动炼钢过程中炉口火焰光谱数据的有效特征提取提供一种快速算法,实现自动炼钢过程的碳含量和温度值的动态预报,以推动智能炼钢进程。本研究采用分波段最小二乘拟合算法对炉口火焰光谱信息的稳定特征进行提取,采用小波分析方法对炉口火焰光谱信息的不稳定特征进行提取,构建了时间、光谱数据主特征、累计耗氧量、碳含量、温度值等一一对应的样本集,借助极限学习机拟合算法构建了不同条件下的自动炼钢过程的碳含量和温度动态预报模型,通过预测误差矩阵、炼钢初始条件实现了基于支持向量机的动态预报模型分类,为不同条件下的样本提供最优的动态预报模型。研究结果表明：应用分波段最小二乘拟合算法和小波分析算法提取的炉口火焰光谱信息的稳定特征和不稳定特征,可以很好地反映全光谱信息;基于绝对误差设计交叉实验,得到的样本类别、预测模型类别和样本初始条件之间呈现出了一致性;光谱信息数据挖掘可以为自动炼钢过程中碳含量和温度值的动态预报进行修正标定,为炼钢终点的控制提供支持。
In the process of automatic steelmaking,in order to provide a fast algorithm for effective feature extraction of furnace mouse flame spectrum data,realize the dynamic prediction of carbon content and temperature value and promote the intelligent steelmaking process,in this study,the sub-band least squares fitting algorithm is used to extract the stable characteristics of the furnace mouse flame spectrum information,the wavelet analysis method is used to extract the unstable characteristics of the furnace mouse flame spectrum information.Then,the one by one corresponding sample sets of time,the main characteristics of spectral data,cumulative oxygen consumption,carbon content and temperature value are constructed; with the help of extreme learning machine fitting algorithm,the dynamic prediction models of carbon content and temperature value under different conditions are constructed; through the prediction error matrix and the initial conditions of steelmaking,the dynamic forecasting model classification based on support vector machine is realized,and the optimal dynamic prediction models are provided for the samples under different conditions.The study results show that the stable characteristics and unstable characteristics of the furnace mouse flame spectrum information extracted with the sub-band least squares fitting algorithm and wavelet analysis algorithm can reflect the full spectrum information nicely.Cross experiment was designed based on the absolute error,the consistency among the sample category,predicted model category and initial condition of the samples is shown; Spectral information data mining can be used to carry out the modified calibration for the dynamic prediction of carbon content and temperature value in the automatic steelmaking process,and provide support for the steelmaking endpoint control.
作者
张彩军
韩阳
何世宇
杨爱民
常锦才
Zhang Caijun;Han Yang;He Shiyu;Yang Aimin;Chang Jincai(College of MetaUurgy and Energy Source, North China University of Science and Technology, Tangshan 063210, China;College of Science, North China University of Science and Technology, Tangshan 063210, China;Yisheng College, North China University of Science and Technology, Tangshan 063210, China)
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2018年第1期24-33,共10页
Chinese Journal of Scientific Instrument
基金
国家自然科学基金(51674121,51474089)
河北省硕士研究生创新(CXZZSS2017071)项目资助
关键词
转炉炼钢
光谱特征
小波分析
超限学习机
支持向量机
converter steelmaking
spectral characteristic
wavelet analysis
extreme learning machine
support vector machine（SVM）
