基于组合训练方法的RBFNN转炉炼钢静态模型  

A Static Model of Converter Steelmaking Using RBFNN Based on Combined Methods

作　　者：朱亚萍[1] 王文龙[1] 徐峥[2] 徐生林[1] 

机构地区：[1]杭州电子科技大学自动化学院,浙江杭州310018 [2]杭州市七格污水处理厂工程建设指挥部,浙江杭州310019

出　　处：《杭州电子科技大学学报（自然科学版）》2011年第3期62-65,共4页Journal of Hangzhou Dianzi University：Natural Sciences

基　　金：省公益技术应用研究资助项目(C31016)

摘　　要：为了提高转炉炼钢终点碳含量和温度的预报命中率,该文采用径向基神经网络建立转炉炼钢静态模型。量子微粒群优化算法具体较好的全局搜索能力,而梯度下降法有较好的局部搜索能力,为了能够发挥这两种算法的优势,该文提出了一种组合训练方法,用来训练径向基神经网络。并通过对某炼钢厂的历史数据进行仿真实验,比较组合训练方法与非组合训练方法的学习性能,研究结果表明该研究提高了转炉终点C含量和温度的预测精度。In order to increase the hit rate of the BOF endpoint,a static model of converter steelmaking was established using radial basis function neural network.Quantum-behaved particle swarm optimization has better global search capability,but gradient descent method has better local search capability.To be able to play the advantages of the two algorithms,a combination of training methods was proposed,to train the radial basis function neural network.Stimulation was done based on the historical data of a steel plant,so as to compare the combination method and non-combination methods.The results indicate that the analysis improves prediction accuracy of the converter end C content and temperature.

关 键 词：径向基神经网络 转炉炼钢 梯度下降法 量子微粒群优化算法 

分 类 号：TF721[冶金工程—钢铁冶金]