摘要
炼钢过程中Cr的加入是一个逐步调整的过程,由于大量随机因素存在,使得Cr含量的测量值低于钢水中Cr的实际含量,从而导致过多Cr加入,成品钢中Cr含量偏高。为此,提出用RBF(RadialBasisFunction)神经网络模型来估计钢水中Cr含量,寻求一组过程变量,通过主元分析降低输入维数,训练神经网络,对钢水中Cr含量进行有效估计。该方法可以一次性调整Cr的含量到要求范围内。
Describes a neural network approach that combines principal component analysis (PCA) with a radial basis function (RBF) network for prediction and control of the chrome content in steel making. Out-of-specification chrome steel is expensive for the manufacturer. The relationship between input parameters and chrome content of the finished steel is non-linear, and not all input parameters can be controlled. PCA is used to determine the important input parameters, decreasing the dimensionality of the input space and showing which parameters must be most closely controlled in the production process. The RBF network is then trained using the important input parameters to predict the final chrome content of the steel. The network has shown excellent prediction accuracy and can be applied in the production process to assist in the determination of appropriate input parameters.
出处
《山东冶金》
CAS
2004年第2期45-47,共3页
Shandong Metallurgy
关键词
炼钢
铬
主元分析
估计模型
RBF神经网络
钢水
steel making
chrome content
radial basis function network
principal component analysis
prediction
