Experimental Research On Coal-rock Identification Method Based On Visible-near Infrared Spectroscopy

In this paper,the coal and rock samples retrieved from the Huainan Xieqiao mine and Paner mine are taken as the research object.The reflectance spectrum curve of the sample is collected by a ground spectrometer,and the sample oxide content,carbon,hydrogen,nitrogen,sulfur,moisture,and ash are also detected.,Volatile matter and fixed carbon content,using the reflectance spectral data of the sample and the content of the sample component as independent variables,and the sample type as the dependent variable to establish a coal and rock classification recognition model to classify coal and rock.This paper mainly uses three models,which are principal component analysis combined with support vector machine(PCA-SVM),principal component analysis combined with BP neural network(PCA-BP)model and kernel principal component analysis combined with support vector machine(KPCA-SVM)model.The specific research conclusions are as follows:1.Due to the low reflectivity of coal and rock blocks,this experiment crushed and sieved the sample,and divided it into five intervals(less than 0.15mm,0.15-0.25mm,0.25-0.5mm,0.5-1mm and 1-3mm),by analyzing the spectral curves of the five particle size intervals,select the spectral data of the particle size interval(less than 0.15mm)that has a significant effect on the sample reflectance amplification.2.Principal component analysis method is used to reduce the dimensionality of the sample reflectance spectral data to obtain the main components that are not related to each other and can represent the spectral characteristics.The obtained main components are respectively established a support vector machine model and a BP neural network model,and randomly The training set and validation set were established to evaluate the classification results.The results showed that the average model recognition rate of the PCA-SVM model was 83.18%and the average recognition rate of the validation set was 45.97%in the 20 groups of random modeling;PCA-BP model The average recognition rate is 46.11%.The recognition rate of PCA-BP model is slightly higher than that of PCA-SVM model.3.Adopt the nuclear principal component analysis method to reduce the dimensionality of the sample reflectance spectrum data,establish the principal vector machine model of the obtained principal component,and verify the model of 20 groups of the same samples as the PCA-SVM model.The results show that KPCA-The modeling recognition rate of the SVM model is up to 100%,the lowest recognition rate is 85%,the average recognition rate is 95.5%,the highest recognition rate of the verification model is 100%,the lowest is 66.67%,and the average recognition rate is about 90.56%;KPCA-The classification and recognition rate of the SVM model is significantly higher than the PCA-SVM model and the PCA-BP model.4.Establish PCA-SVM,PCA-BP and KPCA-SVM models based on the content of sample components for model classification.The results show that the average recognition rate of PCA-SVM model modeling is 83.75%.The average recognition rate of the verification model is 50.56%;the average recognition rate of the PCA-BP model is 46.11%;the average recognition rate of the KPCA-SVM modeling accuracy is 98.5%.The average recognition rate of the verified model is about 95%.The recognition rate of the PCA-SVM model is slightly higher than the PCA-BP model;the classification recognition rate of the KPCA-SVM model is significantly higher than the PCA-SVM model and the PCA-BP model.Figure[38]Table[13]Reference[79]...