摘要
转炉炼钢控制目标是终点温度和碳含量,由于炉温过高,无法在线连续测量.用传统的机理模型建立的终点温度和碳含量模型不够精确.基于RBF神经网络任意逼近函数能力,隐层中心采用最近邻聚类算法,避开K-均值法依赖于聚类中心的初始位置,易陷入局部极小点的缺点.权值调整采用带加权因子的递推最小二乘算法,建立基于RBF神经网络的转炉炼钢终点温度及碳含量的预报模型,并结合某钢铁企业一座180 t转炉的实际数据进行模型验证研究.结果表明,预报精度高于传统的机理模型及BP模型.
The aim of basic oxygen furnace（BOF） steel making endpoint control is the temperature and the carbon content. Because of the high smelting temperature, it is difficult to take measurement accurately and in time. The model is not accurate with the traditional method. On the basis of radial basis fun ction （RBF） approach ability, the nearest neighbor algorithm was used to adjust centers and avoids the disadvantage of the K-means, which relies on the initial central position and is possible to enter local minimum point. The recursive least square method was used to calculate the output weights of middle layers. An RBF network model of endpoint temperature and the carbon content was set up. The practical data of a 180 t converter were simulated. The results show that the precision is higher than that based on the traditional method and backpropagation neural network.
出处
《沈阳工业大学学报》
EI
CAS
2006年第4期405-408,共4页
Journal of Shenyang University of Technology
基金
辽宁省教育厅基金资助项目(202063296)
关键词
转炉炼钢
终点控制
RBF神经网络
最近邻聚类
K-均值聚类
递推最小二乘法
BOF steel making
endpoint control
radial basis function
the nearest neighbors clustering algorithm
K-means clustering algorithm
recursive least square
