烧结配矿乏信息灰自助神经网络特征参数估计
来源期刊：中南大学学报(自然科学版)2012年第12期
论文作者：刘代飞 李军 袁礼顺
文章页码：4917 - 4922
关键词：烧结配矿；乏信息分析；人工神经网络模型；Bootstrap方法；蒙特卡罗模拟
Key words：sinter ore match; poor information analysis; artificial neural networks model; Bootstrap method; Monte Carlo simulation
摘 要：采用BP神经网络建立铁矿烧结评价函数分析烧结配矿特征信息。由中和矿的化学成分、黏附粒子及成核粒子共13个参数构成BP模型输入，烧结矿转鼓强度和烧结速度为模型输出。由100组训练和50组测试数据通过建模确定BP模型隐层节点为27。采用Bootstrap方法处理正交试验数据，采用灰分析提取数据动态信息。提出一种基于人工神经网络模型的参数估计概率求解策略，其中系统整体和动态信息由神经网络模型综合，参数特征值由蒙特卡罗模拟结合概率计算获得。估计和模拟结果为烧结配矿乏信息分析提供一种有效途径。
Abstract: An iron ore sinter evaluating function based on back propagation neural networks (BP) model was adopted to express and analyze complex sinter parameters information. The basic BP model consists of 13 input nodes, 27 hidden nodes and 2 output nodes. The input parameters included essential chemical composition parameters, quantity of conglutination particle and nucleolus particle of mixed ores; the output parameters consisted of sintering velocity and tumbler strength; and the hidden node was obtained through BP modeling with 100 groups of data for training and 50 groups of data for testing. Static information of experiments data was processed by Bootstrap method and change tendency information of process data was analyzed by grey analysis method. A kind of parameter estimation scheme was proposed based on BP model and statistics probability calculation. Sinter general and dynamic information was synthesized by the neural networks models, and estimation values of parameters within the ore evaluating functions were deduced by Monte Carlo simulation integrated statistics probability calculation. The estimation and simulation results provide an effective analysis means for sinter poor information.
刘代飞1，李军2，袁礼顺3
(1. 长沙理工大学 能源与动力工程学院，湖南 长沙，410114；
2. 武汉钢铁股份有限公司烧结厂，湖北 武汉，430080；
3. 中南大学 资源加工与生物工程学院，湖南 长沙，410083)
摘 要：采用BP神经网络建立铁矿烧结评价函数分析烧结配矿特征信息。由中和矿的化学成分、黏附粒子及成核粒子共13个参数构成BP模型输入，烧结矿转鼓强度和烧结速度为模型输出。由100组训练和50组测试数据通过建模确定BP模型隐层节点为27。采用Bootstrap方法处理正交试验数据，采用灰分析提取数据动态信息。提出一种基于人工神经网络模型的参数估计概率求解策略，其中系统整体和动态信息由神经网络模型综合，参数特征值由蒙特卡罗模拟结合概率计算获得。估计和模拟结果为烧结配矿乏信息分析提供一种有效途径。
关键词：烧结配矿；乏信息分析；人工神经网络模型；Bootstrap方法；蒙特卡罗模拟
LIU Dai-fei1, LI Jun2, YUAN Li-shun3
(1. School of Energy and Power Engineering, Changsha University of Science and Technology, Changsha 410114, China;
2. Sinter Plant of Wuhan Iron and Steel Company Limited, Wuhan 430080, China;
3. School of Minerals Processing & Bioengineering, Central South University, Changsha 410083, China)
Abstract:An iron ore sinter evaluating function based on back propagation neural networks (BP) model was adopted to express and analyze complex sinter parameters information. The basic BP model consists of 13 input nodes, 27 hidden nodes and 2 output nodes. The input parameters included essential chemical composition parameters, quantity of conglutination particle and nucleolus particle of mixed ores; the output parameters consisted of sintering velocity and tumbler strength; and the hidden node was obtained through BP modeling with 100 groups of data for training and 50 groups of data for testing. Static information of experiments data was processed by Bootstrap method and change tendency information of process data was analyzed by grey analysis method. A kind of parameter estimation scheme was proposed based on BP model and statistics probability calculation. Sinter general and dynamic information was synthesized by the neural networks models, and estimation values of parameters within the ore evaluating functions were deduced by Monte Carlo simulation integrated statistics probability calculation. The estimation and simulation results provide an effective analysis means for sinter poor information.
Key words:sinter ore match; poor information analysis; artificial neural networks model; Bootstrap method; Monte Carlo simulation
