可见、近红外光谱和深度学习CNN-ELM算法的煤炭分类

煤是工业的主要能源，煤的品质对工业和环境起决定性作用。在使用煤的过程中，如果不能准确确定煤的品种，有可能对生产效率、环境污染、经济损失等会造成重大的影响。传统的煤分类，主要依靠人工方法和化学分析方法，这些方法的缺点是高成本和耗费时间。如何快速准确确定煤的品质很重要。因此，提出深度学习、极限学习机-ELM算法和可见、红外光谱联合建立煤矿分类模型。首先，从抚顺、伊敏和河南夹津口煤矿区采取不同煤样品，并使用美国Spectra Vista公司的SVC HR-1024地物光谱仪测得光谱数据。然后利用深度学习的卷积神经网络-CNN提取光谱特征，并采用ELM算法对光谱数据建立分类模型。最后，为进一步提高分类精度，引入粒子群算法。通过全新定义惯性权重和加速系数的取值范围来改进粒子群算法，并使用改进粒子群算法优化CNN-ELM网络。实验结果表明，和PCA特征提取方法比较，CNN网络能够更好的提取光谱特征，CNN-ELM分类模型有良好的分类效果；改进ELM分类模型的分类精度高于基础ELM和SVM分类模型。与传统的化学分析方法和人工方法相比，此方法在经济、速度、准确性方面均具有无可比的优势。

Coal Classification Based on Visible,Near-Infrared Spectroscopy and CNN-ELM Algorithm

Coal serves as the main energy in industrial field, the quality of which has a decisive effect on industry and environment. In the using process of coal, if the category of the coal fails to be identified correctly, it will result in great harm to production efficiency, environmental pollution and economical loss. The traditional way of classifying coal mainly depends on artificial classification as well as chemical analysis, which however entails high cost and consumes too much time. Therefore, it becomes more and more important to identify the quality of coal quickly and correctly. Hence, this essay comes up with the idea of combining deep learning, ELM arithmetic and visible, infrared spectra to construct coal classification model. Firstly, we collected different coal samples from Fushun, Yimin and Henan Jiajinkou coal mining area, and used the American Spectra Vista SVC HR-1024 spectrometer for the measurement of the spectral data. Then we used the deep learning of convolutional neural network-CNN to extract spectral characteristics, and adopted ELM arithmetic to construct classification model for spectral data. Finally, in order to further improve the classification accuracy, this article made use of particle swarm optimization algorithm by using a range of newly defined inertia weight and acceleration factor values to improve the particle swarm optimization algorithm. Then, we used the improved particle swarm optimization to optimize CNN-ELM networks. Experimental results from comparison between PCA and CNN network reveal CNN network as a better feature extraction method for the spectrum. The results also show that CNN-ELM classification model has a good classification effect. The improved ELM classification model accuracy is higher than that of the basic ELM and SVM classification model. Compared with the traditional chemical methods and artificial methods, this method has the advantage of being unparalleled in economy, speed and accuracy.