基于深度学习的浮选精矿品位预测方法  

Prediction method of flotation concentrate grade based on deep learning

作　　者：赵玉华 杨文旺 武涛 ZHAO Yu-hua;YANG Wen-wang;WU Tao(BGRIMM Machinery and Automation Technology Co.Ltd.,Beijing 100160,China;BGRIMM Technology Group,Beijing 100160,China)

机构地区：[1]北矿机电科技有限责任公司,北京100160 [2]矿冶科技集团有限公司,北京100160

出　　处：《矿冶》2022年第4期70-74,101,共6页Mining And Metallurgy

摘　　要：矿浆品位是浮选工艺中关键参数之一,其对于指导生产、节约药剂、控制产品质量和提高回收率等方面都起着非常关键的作用。为了在线预测浮选精矿品位,解决荧光分析仪检测滞后的问题,研究出了一种不需要主观提取特征的基于深度学习的精矿品位在线预测模型,模型的输入为浮选泡沫图像序列、原矿品位值和尾矿品位值,输出为精矿品位值,属于回归问题。对比了主干网络分别为VGG-16、ResNet-50和MobileNet-V2时预测结果的差异,实验结果显示VGG-16的预测精度和鲁棒性最好,平均预测精度达到12.48%。The pulp grade is one of the key parameters in the flotation process,which plays a key role in guiding production,saving chemicals,controlling product quality and improving recovery.In order to predict the flotation concentrate grade online and solve the problem of the detection lag of the fluorescence analyzer,a deep learning-based online prediction model of concentrate grade without subjective extraction of features is developed.The value and tailings grade value are input,and the output concentrate grade value is a regression problem.Comparing the differences in the prediction results when the backbone network is VGG-16,ResNet-50 and MobileNet-V2,the experimental results show that VGG-16 has the best prediction accuracy and robustness,with an average prediction accuracy of 12.48%.

关 键 词：图像序列 深度学习 精矿品位 浮选泡沫 

分 类 号：TD952[矿业工程—选矿] TP272[自动化与计算机技术—检测技术与自动化装置]