基于神经网络集成-X射线荧光光谱法的铁矿石中全铁含量测定

Determination of total iron in iron ore based on ensemble neural network-X-ray fluorescence spectrometry

为了探索人工智能在铁矿石品质快速检验中的应用,研究了机器学习算法与化学计量学和X射线荧光光谱仪(XRF)相结合快速测定铁矿石中全铁含量的方法。收集来自于不同产地的,主要物相为赤铁矿、褐铁矿、磁铁矿、针铁矿和多物相混和结构的铁矿石样品共1 098个作为样本集。采用X射线荧光光谱仪对铁矿石样品熔片进行扫描,扫描后的光谱图提取数据点后作为神经网络的输入,以全铁含量作为输出结果。然后依据X射线衍射(XRD)得到的物相结构优化自组织(SOM)网络,并对全部样本的XRF图谱进行分类,对分类后的每一个子集分别采用反向传播(BP)和径向基函数(RBF)网络建立回归子模型,对各子模型的预测结果进行整合,最终建立基于集成神经网络和X射线荧光光谱法的铁矿石中全铁含量预测模型。方法模型建立后,不需要额外标准物质建立校准曲线,能够实现对未知样品的分类和输出全铁含量结果。

In order to explore the application of artificial intelligence in rapid quality test of iron ore,the rapid determination method of total iron content in iron ore was studied by machine learning algorithm,chemometrics and X-ray fluorescence spectrometry(XRF).Total 1098 iron ore samples from different regions(the main phases were hematite,limonite,magnetite,goethite and multi-phase mixture structure)were used as the sample set.The fuse pieces of iron ore samples were scanned by X-ray fluorescence spectrometer.The data points extracted from the spectra were used as the input of neural network,while the content of total ion was used as the output result.Then,the self-organization mapping(SOM)networks were optimized according to the phase structure obtained by X-ray diffraction(XRD).The XRF spectra of all samples were classified.The regression submodel of each subset was established based on back propagation(BP)and radial basis function(RBF)network.The prediction results of each submodel were integrated.Finally the prediction model of total iron content in iron ore based on neural network integration and XRF was established.The model experience certificate test results indicated that the accuracy was up to 89.1%.After model establishment,the additional standard substances were not required to establish calibration curve for the classification and total iron content output of unknown samples.