基于X射线吸收光谱的煤矸石鉴别

Identification of coal gangue based on X-ray absorption spectroscopy

针对纹理识别、图像处理、红外辐射传统方法对于煤矸石的快速鉴别的准确率不高及鉴别时间过长的问题,利用X射线吸收光谱(XAS)的白矸石和煤矸石的快速鉴别,对部分样本进行光谱数据测量,绘制光谱图像,分析不同煤矸石占比混合物的光谱区别所在.进而继续扩大样本空间多次重复测量.经去背景、归一化等方法处理最终得到了160组吸收光谱数据.应用主成分分析,对主成分采用最小二乘法得到线性回归方程,通过方程可以求解得到预测结果.为了避免线性回归效果不佳的情况,还对数据进行了非线性回归分析,根据训练集数据建立径向基神经网络,对比测试集输出和期望值得出误差进而检验精度.结果可见采用RBF神经网络的算法准确度更高,得到的结果平均绝对误差仅0.86%,均方误差仅为0.44%,可以较为精确地完成煤矸石含量预测.

Traditional methods such as texture recognition,image processing,and infrared radiation have low accuracy and long identification time for rapid identification of coal gangue.With the rapid identification of white gangue and coal gangue by X-ray absorption spectroscopy(XAS),the spectral data of some samples were measured,the spectral images were drawn,and the spectral difference of different gangue-occupied mixtures was analyzed.Furthermore,the sample space was expanded and measurements multiple times were repeated.By removing the background,normalization and other methods of processing,160 groups of absorption spectral data were finally obtained.Applying principal component analysis,the least squares method was used to obtain a linear regression equation for the principal components,which can be solved to obtain the predicted results.In order to avoid the poor effect of linear regression,a nonlinear regression analysis on the data was also conducted.A radial basis function(RBF) neural network was built based on the training set data.The test set output and the expected value were compared to derive the error and then the accuracy was checked.The results show that the algorithm using RBF neural network has higher accuracy,with an average absolute error of only 0.86% and a mean square error of only 0.44%,which can accurately predict the content of coal gangue.