由工业发展需求, 针对菱镁矿石矿物含量不同以及分布不均匀而难以判定其品级的情况, 提出一种由近红外光谱技术结合ELM的菱镁矿石品级分类模型。 该模型可以实现菱镁矿石品级的快速分类。 近红外光谱利用菱镁矿中不同种类含H基团对近红外光谱有不同吸收的特性, 用来测定菱镁矿石的成分及其含量, 其操作简便、 不破坏样品、 速度快、 准确高效。 以辽宁省营口市大石桥的菱镁矿石30组为研究对象, 采集菱镁矿石的近红外光谱数据样本30×973。 采用主成分分析(PCA)对其进行降维处理, 以主元贡献率大于99.99%而得到10维的特征变量值。 建立了ELM算法定量分析数学模型, 取20组样本为训练样本(包括6组特级, 14组非特), 其余10组作为测试样本(其中4组特级, 6组非特), ELM算法模型的隐含层节点数选取20。 为了进一步提高分类效果, 提出两种ELM算法模型的改进: 采用循环模式对传统ELM的输入权值和阈值进行寻优的精选ELM和在精选ELM基础上进行集成的集成-精选ELM。 并与用人工方法、 化学方法和BP神经网络模型方法对菱镁矿石样品品级分类作对比。 结果表明: 近红外光谱和ELM菱镁矿石品级分类模型不论在时间上还是成本上, 都具有明显的优势, 且其准确率能够达到90%以上, 为菱镁矿石品级分类提供了一条新的途径。

Due to the needs of industrial development, the different content and uncertain distribution of magnesite mineral lead to great difficulties in o determining its grade, therefore, we propose a combination of near-infrared spectroscopy and the ELM magnesite grade classification model. The model can achieve rapid classification of magnesite grade. Near infrared spectroscopy, considering that different types of H group in magnesite have different absorption degrees to near-infrared spectroscopy, is used to determine the composition and content of magnesite. It is simple, fast, accurate and efficient without destroying the sample. In this paper, we take magnesite 30 group from Yingkou City, Liaoning Province Dashiqiao for the study, collecting their magnesite NIR data samples at 30×973, using principal component analysis (PCA) for data dimensionality reduction process. The main element contribution rate is greater than 99.99% obtained characteristic variables of 10, established quantitative analysis ELM algorithm mathematical model, take 20 groups of samples as the training samples (including 6 super group, 14 groups non), 10 groups of samples for testing samples (including super grade4 groups, 6 groups non), ELM algorithm model hidden layer nodes selection 20. In order to further improve the classification performance, two kinds improved ELM algorithm models are proposed: conduct optimization selection ELM for the traditional ELM input weights and threshold using the circulation patterns and integrate integration-Featured ELM based on Featured ELM. And compare to which use the artificial method, chemical method and BP neural network model approach. The results showed that magnesite grade classification with the near-infrared spectroscopy and ELM model have a distinct advantage regardless of cost or time, and the accuracy rate can reach over 90%, which provides a new way to classify magnesite grade.