基于深度学习的浮选回收率预测建模研究  

Research on Flotation Recovery Rate Modeling Based on Deep Learning

作　　者：赵红宇 何桂春 石岩[1,2] 江长松 吴为波 ZHAO Hongyu;HE Guichun;SHI Yan;JIANG Zhangsong;WU Weibo(Jiangxi Province Key Laboratory of Mining Engineering,Ganzhou 341000,China;School of Resources and Environmental Engineering,Jiangxi University of Science and Technology,Ganzhou 341000,China)

机构地区：[1]江西省矿业工程重点实验室,江西赣州341000 [2]江西理工大学资源与环境工程学院,江西赣州341000

出　　处：《金属矿山》2024年第1期261-268,共8页Metal Mine

基　　金：国家自然科学基金项目(编号:52174249);江西省重点研发计划项目(编号:20203BBGL73231);江西省研究生创新专项资金项目(编号:YC2022-S660)。

摘　　要：针对现有浮选回收率预测模型拟合度不高、预测误差大等问题,以某铜矿实际工况数据为基础,利用箱图和滤波算法对数据进行预处理,采用传统机器学习算法(DT、SVR和RF算法)和深度学习算法(DNN和CNN算法)构建相应浮选回收率预测模型。对5种回收率预测模型的拟合效果、预测效果进行了对比分析,并采用现场数据进行验证。结果表明:传统机器学习算法模型中RF预测精度最佳,±2%误差区域命中率为80.1%,±4%误差区域命中率为93.0%;深度学习模型预测效果均优于传统机器学习算法模型,DNN和CNN预测模型的R~2分别为0.854、0.907,±2%误差区域命中率分别为91.6%、90.6%,±4%误差区域命中率分别为96.6%、98.1%。CNN模型略优于DNN模型,但训练耗时较长,深度学习算法模型中首选DNN模型。研究结果可为浮选回收率实时预测及浮选过程协同优化提供技术支持。Aiming at the problems such as low fitting degree and large prediction error of the existing flotation recovery prediction model,based on the actual working condition data of a copper mine,the box diagram and filtering algorithm were used to pre-process the data,and the corresponding flotation recovery prediction model was constructed by traditional machine learning algorithms(DT,SVR and RF algorithms)and deep learning algorithms(DNN and CNN algorithms).The fitting effect and prediction effect of five recovery prediction models were compared and analyzed,and verified by field data.The results showed that the RF prediction accuracy of the traditional machine learning algorithm model is the best,the error area of±2%is 80.1%,and the error area of±4%is 93.0%.The prediction effect of the deep learning model is better than that of the tradi-tional machine learning algorithm model.The R2 of the DNN and CNN prediction models are 0.854 and 0.907,respectively;the accuracy of the±2%error region is 91.6%and 90.6%,respectively;The accuracy of the±4%error region is 96.6%and 98.1%,respectively.The CNN model is slightly better than the DNN model,but the training time is longer,thus the DNN mod-el is the first choice in the deep learning algorithm model.The research results could provide technical support for real-time prediction of flotation recovery rate and collaborative optimization of flotation process.

关 键 词：浮选回收率 机器学习 数据预处理 深度学习 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TD923[自动化与计算机技术—控制科学与工程]