基于选矿生产数据的矿石可磨性智能识别算法  

Intelligent identification algorithm of ores grindability based on mineral processing production data

作　　者：朱佼佼 李宗平[1] 师本敬 吴革雄 文武 ZHU Jiaojiao;LI Zongping;SHI Benjing;WU Gexiong;WEN Wu(Zhongye Changtian International Engineering Co.,Ltd.,Changsha 410205,Hunan,China)

机构地区：[1]中冶长天国际工程有限责任公司,湖南长沙410205

出　　处：《烧结球团》2024年第4期83-92,共10页Sintering and Pelletizing

基　　金：五矿科技专项计划项目(2020ZXC03)。

摘　　要：本文以在线识别矿石可磨度为研究目标,通过详细分析磨矿过程参数间的相互影响关系,得出磨矿过程运行指标的变化能从一定程度上反应矿石可磨度变化的结论。针对工业数据维度高、耦合性强等问题,在着重研究了KPLSR和SAE两种可提取关键特征算法机理的基础上,以实际采集的磨矿过程数据为样本,对多算法进行横纵对比,最终验证了深度学习算法SAE在工业复杂数据分析中的优势。并以测试试验中性能最佳的算法结构作为矿石可磨性智能识别系统的在线识别模型,该模型在工业化应用中绝对值误差平均可达0.61 kW·h/t,平均误差率约为4.79%,可满足现场生产的需求。The research goal is to identify the ore grindability online in this paper.Through a detailed analysis of the interaction between the grinding process parameters,it is concluded that the change of the grinding process operation index can reflect the change of ore grindability to a certain extent.In order to solve the problems of high dimensionality and strong coupling of industrial data,on the basis of focusing on the mechanism of KPLSR and SAE two algorithms that can extract key features,the actual collected grinding process data is used as a sample to compare multiple algorithms horizontally and vertically,and finally the advantages of deep learning algorithm SAE in industrial complex data analysis are verified.The algorithm structure with the best performance in the test is used as the online recognition model of the ore grindability intelligent identification system,and the average absolute error of the model can reach 0.61 kW·h/t in industrial applications,and the average error rate is about 4.79%,which can meet the needs of on-site production.

关 键 词：矿石 可磨度 磨矿过程 数据驱动 软测量技术 深度学习算法 

分 类 号：TD921[矿业工程—选矿] TP29[自动化与计算机技术—检测技术与自动化装置]