基于XRT感应特性及BP神经网络的某钨矿石品位预测模型  

A tungsten ore grade prediction model based on XRT induction characteristics and BP neural network

作　　者：李思佑 李丽匣[1] 张宏亮 徐阳 张依然 张晨 晏丽鑫 LI Siyou;LI Lixia;ZHANG Hongliang;XU Yang;ZHANG Yiran;ZHANG Chen;YAN Lixin(College of Resources and Civil Engineering,Northeastern University,Shenyang 110819,China;Ganzhou Best Friend Technology Co.Ltd.,Ganzhou 341000,China;College of Mining Engineering,University of British Columbia,Vancouver V6T 1Z4,Canada;Beijing Research Institute of Chemical Engineering and Metallurgy,Beijing 101149,China)

机构地区：[1]东北大学资源与土木工程学院,辽宁沈阳110819 [2]赣州好朋友科技有限公司,江西赣州341000 [3]加拿大英属哥伦比亚大学矿业工程学院,加拿大温哥华,V6T 1Z4 [4]核工业北京化工冶金研究院,北京101149

出　　处：《中南大学学报（自然科学版）》2024年第11期4124-4134,共11页Journal of Central South University:Science and Technology

基　　金：国家重点研发计划项目(2022YFC2903802)。

摘　　要：为了实现对某钨矿石的XRT(X射线透射)高效预选,建立XRT图像与矿石品位的相关模型。首先,以某钨矿石作为试验样本进行X射线透视成像,并基于MATLAB软件获得XRT灰度图像,建立灰度直方图,统计图像的灰度分布;其次,分别建立基于BP、GA-BP神经网络的WO_(3)品位预测模型,并对钨矿石样本的模型进行训练和测试,比较2种预测模型的预测效果;最后,基于模型的精度与泛化能力,确定合适的钨矿石品位预测模型及优化方法。研究结果表明:不同品位矿石XRT灰度图像的灰度分布存在明显差异,图像的灰度分布与矿石品位之间具有高度相关性;钨矿石品位越高,像素灰度级在低灰度区间的占比越大,可通过XRT图像灰度分布建立矿石品位的预测模型;使用遗传算法对BP神经网络进行优化可以取得显著效果,基于GA-BP神经网络的预测模型能够获得更大的决定系数和更小的误差,具有更高的预测精度和更强的泛化能力,可以更好地预测矿石WO_(3)品位;在小样本情况下,GA-BP神经网络预测模型对于WO_(3)品位预测具有可行性与有效性。In order to achieve efficient XRT preselection for a tungsten ore,a model between XRT image and ore grade was established.Firstly,a tungsten ore was taken as a test sample for X-ray perspective imaging.Using MATLAB software,XRT grayscale image of the ore was generated,gray histogram was established and grayscale distribution of the image was counted.Secondly,based on BP and GA-BP neural networks,WO_(3) grade prediction models were established respectively,and the tungsten ore samples were used for training and testing.Coefficient of determination,root mean square error,mean absolute error,and mean deviation error of two prediction models were analyzed.Finally,based on the accuracy and generalization ability of the two prediction models,the appropriate tungsten ore grade prediction model and optimization method were determined.The results show that the grayscale distribution of XRT grayscale images of ores with different grades is obviously different.The grayscale distribution obtained from the images is highly correlated with the ore grade.The higher the grade of tungsten ore,the greater the proportion of pixel grayscale level at the low grayscale interval.The prediction model of ore grade can be established by the grayscale distribution of XRT images.The prediction model based on GA-BP neural network can obtain larger coefficient of determination and smaller error.It has higher prediction accuracy and stronger generalization ability,which can better predict WO_(3) grade of tungsten ore.In the case of small samples,GA-BP neural network prediction model is feasible and effective for WO_(3) grade prediction.

关 键 词：品位预测模型 XRT灰度图像 BP神经网络 遗传算法 

分 类 号：TD9[矿业工程—选矿]