摘要
高炉炼铁是一个物理化学反应复杂、多相多场耦合的大滞后、非线性动态密闭系统,其关键质量指标―铁水温度、Si含量、P含量、S含量难以直接在线检测,且离线化验过程滞后严重.针对这一实际工程难题,本文提出一种基于均方根误差概率加权集成学习建模算法,用于高炉多元铁水质量的预测建模.首先,为了提高建模数据质量,对高炉原始数据进行时间粒度的统一、数据归一化等数据预处理操作;为了提高建模效率和降低计算复杂度,采用灰色关联分析法提取与多元铁水质量指标关联度最强的关键变量作为建模输入变量.然后,为了提高建模的精度,提出一种均方根误差概率加权集成随机权神经网络(RVFLNs)算法.该算法采用具有快速建模速度的RVFLNs为子模型,使用核密度估计方法估计出子模型的均方根误差概率密度函数曲线,进而求出每个子模型的均方根误差概率并作为自身权重进行加权求和,得到最终的均方根误差加权集成RVFLNs模型.最后,数值仿真验证和工业试验表明:所提算法能够根据实时输入数据的变化对多元铁水质量进行快速准确的预测.
Blast furnace ironmaking is a large hysteresis and nonlinear dynamic closed system with complex physical and chemical reactions and multi-phase and multi-field coupling. Its key quality indicators―molten iron temperature,Si content, P content and S content are difficult to be directly detected online. There is serious hysteresis in the offline testing process. Aiming at solving this engineering problem, this paper proposes an integrated learning modeling algorithm based on root-mean-square error probability weighting, which is for predictive modeling of multivariate molten iron in blast furnace. Firstly, in order to improve the quality of the modeling data, data pre-processing such as time granularity unification and data normalization;in order to improve the modeling efficiency and reduce the computational complexity,the gray correlation analysis method is used to extract the key variables with the strongest correlation with the quality index of multivariate molten iron as the input variables of the modeling. Secondly, in order to improve the accuracy of modeling,a root mean square error probability weighted integrated RVFLNs algorithm is proposed. The algorithm uses random vector functional-link networks as a sub-model. The kernel density estimation method is used to estimate the root mean square error probability density curve of these sub-models. The root mean square error probability of each submodel is weighted and summed as its own weight, and the final root mean square error weighted integrated RVFLNs model is obtained. Finally,numerical simulation and industrial experiments show that the proposed algorithm can quickly and accurately predict the quality of multivariate molten iron based on changes in real-time input data.
作者
刘进进
周平
温亮
LIU Jin-jin;ZHOU Ping;WEN Liang(State Key Laboratory of Synthetical Automation for Process Industries,Northeastern University,Shenyang Liaoning 110819,China)
出处
《控制理论与应用》
EI
CAS
CSCD
北大核心
2020年第5期987-998,共12页
Control Theory & Applications
基金
国家自然科学基金项目(61890934,61473064,61790572)
中央高校基本科研业务项目(N180802003)资助.
关键词
高炉炼铁
多元铁水质量
灰色关联分析法
核密度估计方法
均方根误差概率加权
数据驱动建模
blast furnace ironmaking
multivariate molten iron quality
grey correlation analysis method
kernel density estimation method
root mean square error probability weighted
data driven modeling
