可见光-近红外光谱的矽卡岩型铁矿反演模型

铁矿资源是我国国民经济基础产业中的重要组成要素，在我国经济发展中有举足轻重的地位。铁矿品位的检定效率对铁矿石开采效率有重大影响。目前，铁矿石品位的化学分析检定法，不仅存在成本较高，化验周期长的问题，更主要的是其无法实现铁矿品位原位测定，相对配矿流程存在滞后效应，无法有效降低矿石开采的损失贫化率；基于可见光-近红外光谱分析的铁矿品位原位测定技术是解决这一问题的有效途径。以225个红岭矽卡岩型铁矿测试样本的可见光-近红外光谱数据及化学分析数据为数据源，首先对原始数据进行了平滑处理，并分析了矽卡岩型铁矿可见光-近红外光谱特征，然后利用倒数对数、多元散射校正（MSC）两种预处理方法对平滑后的光谱数据进行处理，再分别以主成分分析法（PCA）、遗传算法（GA）两种降维算法对预处理前后的光谱数据进行了处理，获取了六种不同预处理组合算法处理后的数据源。其中以PCA降维算法所降维数分别为3维、3维、7维；以GA降维算法所降维数分别为477维、489维、509维。最后基于随机森林（RF）和极限学习机（ELM）建立了矽卡岩型矿石金属铁品位的定量反演模型，以决定系数（R2）、均方根误差（RMSE）和平均相对误差（MRE）三个指标分别对模型的稳定性、精确度、可信度进行评价。结果表明，经MSC处理及PCA降维后的数据基于ELM算法建立的定量反演模型效果最优，其R2可达0.99、RMSE为0.005 7、MRE为2.0%，该方法所建模型对红岭矽卡岩型铁矿品位反演精度有明显的提升。对矽卡岩铁矿品位的实时、快速分析提供了一种有效的方法，对实现矽卡岩型铁矿的高效开采具有重要的现实意义。

Quantitative Inversion Model Based on the Visible and Near-Infrared Spectrum for Skarn-Type Iron Ore

Iron ore resources are an important component of the basic industries of China’s national economy and play a pivotal role in China’s economic development.In particular,the efficiency of iron ore grade determination has a significant impact on the efficiency of iron ore mining.At present,the method of iron ore grade determination is mainly based on chemical analysis.However,it not only has the problems of high cost and long assay cycle but also cannot achieve the in-situ determination of iron ore grade,which has a lag effect relative to the ore allocation process and cannot effectively reduce the loss depletion rate of ore mining,so the in-situ determination of iron ore grade based on visible and near-infrared spectral analysis is an effective way to solve this problem.This paper uses visible and near-infrared spectral data and chemical analysis data from 225 test samples of Hongling Skarn-Type iron ores as data sources.First,the original data were smoothed and analyzed for visible and near-infrared spectral characteristics of Skarn-Type iron ores,and then the smoothed spectral data were processed by using two pre-processing methods,including logarithm of reciprocal and multiple scattering correction(MSC).-Before and after pre-processing,the spectral data were processed using two-dimensionality reduction algorithms,including genetic algorithm(GA)and principal component analysis(PCA),and obtain the data sources were processed by six different pre-processing combination algorithms.The PCA dimensionality reduction algorithm was used to reduce the dimensionality of the spectral data before and after the pre-processing,and the dimensionality reduced were 3,3 and 7 dimensions respectively;The GA dimensionality reduction algorithm was used to reduce the dimensionality of the spectral data before and after the pre-processing,and the dimensionality reduced were 477,489 and 509 dimensions respectively.Finally,based on Random Forest(RF)and Extreme Learning Machine(ELM),a quantitative inversion model of iron grades in skarn-type ores was established,and the stability,accuracy and credibility of the model were evaluated in terms of coefficient of determination(R²),root means square error(RMSE)and mean relative error(MRE).The results show that the quantitative inversion model based on the ELM algorithm,using MSC-processed and PCA-dimensioned data,is the most effective,with R² of 0.99,RMSE of 0.0057 and MRE of 2.0%.The accuracy of the model for inversion of HongLing skarn-type iron ore grades has been significantly improved.This research provides an effective method for the real-time and rapid analysis of skarn-type iron ore grade,which is of great practical importance for efficient skarn-type iron ore mining.