基于深度森林的磨矿粒度软测量模型的实现  

Realization of Soft Measurement Model of Grinding Particle Size Based on Deep Forest

作　　者：秦楠 骆俊 程小舟 刘自杰 孙晨 张思涵 徐瑞玲 陶陶 QIN Nan;LUO Jun;CHENG Xiaozhou;LIU Zijie;SUN Chen;ZHANG Sihan;XU Ruiling;TAO Tao(Sinosteel Maanshan General Institute of Mining Research Co.,Ltd.;School of Computer Science and Technology,Anhui University of Technology;Sinosteel Mining Institute(Maanshan)Intelligent Emergency Technology Co.,Ltd.)

机构地区：[1]中钢集团马鞍山矿山研究总院股份有限公司,安徽省马鞍山市243000 [2]安徽工业大学计算机科学与技术学院 [3]中钢矿院(马鞍山)智能应急科技有限公司

出　　处：《现代矿业》2023年第11期108-111,140,共5页Modern Mining

基　　金：安徽省重点研究与开发计划项目(编号:202104a05020025)。

摘　　要：为了掌握矿业生产中溢流粒度的实时情况,针对磨矿过程中粒度分析仪表易出现故障和损坏的问题,进行了相关实际考察,提出了一种改进深度森林方法。通过选取某选矿厂的真实生产数据集作为样本,将改进深度森林算法和决策树、随机森林、传统深度森林算法进行了对比。结果表明:在训练集上,改进深度森林算法损失较决策树降低了23.1%,较随机森林降低了15.4%,较传统深度森林降低了9.51%;在测试集上,改进深度森林算法损失较决策树降低了23.5%,较随机森林降低了12.8%,较传统深度森林降低了6.4%,并且对比数学预测模型和主流的序列预测方法,证明改进深度森林算法的预测精度明显高于其他方法,能有效提高水力旋流器溢流粒度预测的准确性,对选矿工业的发展具有重要的理论意义与实用价值。In order to grasp the real-time situation of overflow particle size in mining production,in view of the problem that the particle size analyzer is prone to failure and damage in the grinding process,the relevant practical investigation was carried out,and an improved deep forest method was proposed.By selecting the real production data set of a concentrator as a sample,the improved deep forest algorithm is compared with the decision tree,random forest and traditional deep forest algorithm.The results show that on the training set,the loss of the improved deep forest algorithm is 23.1%lower than that of the decision tree,15.4%lower than that of the random forest,and 9.51%lower than that of the traditional deep forest.On the test set,the loss of the improved deep forest algorithm is 23.5%lower than that of the decision tree,12.8%lower than that of the random forest,and 6.4%lower than that of the traditional deep forest.Compared with the mathematical prediction model and the mainstream sequence prediction method,it is proved that the prediction accuracy of the improved deep forest algorithm is significantly higher than that of other methods,which can effectively improve the accuracy of hydrocyclone overflow particle size prediction,and has important theoretical significance and practical value for the development of mineral processing industry.

关 键 词：磨矿分级 溢流粒度 深度森林 回归预测 

分 类 号：TD921.4[矿业工程—选矿]