基于I-GWO-BP神经网络的紫金山铜矿浮选回收率预测研究  

Flotation Recovery Prediction of Zijinshan Copper Ore Based on I−GWO−BP Neural Network

作　　者：夏永涛 马英强[1,2] 印万忠 衷水平[1,2] 鲁军 张德文[3] 詹殷权[1] XIA Yongtao;MA Yingqiang;YIN Wanzhong;ZHONG Shuiping;LU Jun;ZHANG Dewen;ZHANYinquan(Zijin School of Geology and Mining,Fuzhou University,Fuzhou 350108,China;FujianKey Laboratory of Green Extraction and High−value Utilization of New Energy Metals,Fuzhou 350108,China;Zijin Mining Group Co.,LTD.,Shanghang 364200,China)

机构地区：[1]福州大学紫金地质与矿业学院,福建福州350108 [2]福建省新能源金属绿色提取与高值利用重点实验室,福建福州350108 [3]紫金矿业集团股份有限公司,福建上杭364200

出　　处：《矿产保护与利用》2023年第3期51-59,共9页Conservation and Utilization of Mineral Resources

基　　金：国家自然科学基金项目(51804081);福建省自然科学基金项目(2019J01253);福州大学本科生科研训练计划项目(28212)。

摘　　要：为克服传统测量浮选回收率方式存在的低效率、滞后性等问题,结合紫金山硫化铜矿浮选厂生产情况,采用基于MI(Mutual Information)互信息法对选厂原矿品位、丁铵黑药用量等浮选条件因子进行特征选择,在此基础上,建立了基于BP(Back Propagation)、GWO-BP(Grey Wolf Optimizer-Back Propagation)、I-GWO-BP(Improved-Grey Wolf Optimizer-Back Propagation)的三种浮选回收率预测模型,并选取紫金山硫化铜矿浮选车间生产数据进行神经网络训练与验证试验,分析了浮选回收率预测模型的准确性。结果表明:相较于基于BP、GWO-BP的浮选回收率预测模型而言,基于I-GWO-BP的浮选回收率预测模型具有更大的相关系数和更小的均方误差根,说明该模型泛化拟合能力更强,对浮选回收率的预测值在很大程度上逼近于真实值,预测精度更高。本研究结果可为实现浮选回收率高效、准确、自动的在线预测技术开发提供支持。The traditional method of measuring flotation recovery has some problems,such as low efficiency and hysteresis.Combined with the flotation plant production of Zijinshan sulfide copper ore,characteristics selection of flotation condition factors such as raw ore grade and dosage of ammonium dibutyl dithiophosphate was carried out based on MI(Mutual Information)method.On this basis,three prediction models of flotation recovery were established based on BP(Back Propagation)GWO−BP(Grey Wolf Optimizer−Back Propagation)and I−GWO−BP(Improved−Grey Wolf Optimizer−Back Propagation).The flotation workshop production data of Zijinshan sulfide copper ore were selected for neural network training and verification test,and the accuracy of the flotation recovery prediction model was analyzed.The results showed that compared with BP and GGO−BP,the flotation recovery prediction model based on I−GWO−BP had a root mean squared error and a correlation coefficient and the predicted value of flotation recovery was the closest to the true value,and the generalization ability of the network was significantly stronger.The results of this study can support the development of efficient,accurate and automatic online prediction techniques for flotation recovery.

关 键 词：浮选回收率 互信息法 BP神经网络 改进的灰狼算法 

分 类 号：TD952.1[矿业工程—选矿]