铜闪速熔炼神经网络模型的建立  

Developing of the Copper Flash Smelting Model based on Neural Network

作　　者：曾青云[1] 汪金良[1] 

机构地区：[1]南方冶金学院材料与化学工程学院,江西赣州341000

出　　处：《南方冶金学院学报》2003年第5期15-18,40,共5页Journal of Southern Institute of Metallurgy

基　　金：江西省自然科学基金资助项目(0250026).

摘　　要：在分析铜闪速熔炼控制策略和贵溪冶炼厂闪速炼铜生产实践数据的基础上,基于神经网络技术分别建立了网络结构为8-16-1的冰铜品位模型、8-20-1的冰铜温度模型和8-20-1的渣中Fe/SiO2模型,并采用了遗传算法GA与前向神经网络算法BP相结合的网络训练方法对其进行了仿真计算,结果显示其检验样本的最大误差分别为0.57、0.84和0.07,最大相对误差分别为0.99%、0.07%和5.1%,均方误差分别为0.37、0.66和0.04,表明所建立模型的预测值与生产实践数据基本吻合,该模型可以应用于铜闪速熔炼过程的在线优化控制.On the basis of the analyses of the copper flash smelting controlling policy and the prac-tical product data of copper flash smelting process in Guixi Smelter,the matte grade model,the matte temperature model and the Fe/SiO 2 ratio in slags model are developed based on neural net-work and the combination of Genetic Algorithms and Back Propagation Algorithm.Those models' respective net structures are8-16-1,8-20-1and8-20-1,the biggest error of the check samples are0.57,0.84and0.07,and the biggest relative error are0.0099,0.0007and0.051,and the aver-age square error are0.37,0.66and0.04.Those results show that the simulative results accord to the practical data.Thus,the models can be used in the optimizing control for copper flash smelting process.

关 键 词：铜 闪速熔炼 神经网络 仿真 

分 类 号：TF811[冶金工程—有色金属冶金]