基于RBF神经网络在转炉炼钢终点预报中的应用研究

Research on application of RBF neural network in endpoint prediction of converter steelmaking

转炉炼钢控制目标是对终点温度和含碳量进行预测。由于我国转炉炼钢自动化控制水平的限制,特别是动态控制水平不够高,因此需要基于RBF神经网络建立终点预报模型。其基本思路为:基于RBF神经网络局部逼近网络的特性之上,采用k-均值聚类算法确定隐藏层的中心,权值调整采用递推最小二乘法,建立基于RBF神经网络在转炉炼钢终点预报的模型。最后结合实际数据进行模型的仿真研究。结果表明经RBF神经网络预测模型的实时训练,提高了终点预报的精度。

Converter steelmaking control target is to predict the temperature and carbon content. Due to the limit of the level of automation control, especially the dynamic control level is not high enough, therefore, it is necessary to establish the endpoint prediction model based on RBF neural network. The basic idea is based on RBF neural network locally close to characteristic of network, adopt k-mean clustering algorithm to determine the center of hidden layer, weight adjustment adopts least square method to establish model for predicting the end point of BoF steelmaking based on RBF neural network. Finally, the simulation research is carried out based on the actual data, the results show that the accuracy of endpoint prediction is improved through practical training of predicting model based on RBF neural network.