Research Of Quantitative Analysis Methods On Laser-induced Breakdown Spectroscopy Coal Analyzer

Coal still occupies a major position in China's current energy market.Fast online coal quality detection methods have great significance to the production efficiency of application scenarios such as coal-fired power plants.Offline coal quality detection methods using traditional methods have become increasingly disadvantageous due to tedious procedures and hard to satisfy the needs of industrial production.The laser-induced breakdown spectroscopy(LIBS)detection technology is very suitable for the needs of online testing of coal quality due to its advantages such as fast testing,support for in-situ detection,and multi-element simultaneous analysis.Many mature products have been derived from soil and water quality testing.However,the development of domestic LIBS instruments in the field of coal quality testing has just begun,which is a cutting-edge research with great application value.This paper addresses the limitations of current widely used standard curve methods,multiple linear regression methods and other quantitative analysis methods which are difficult to analyze and detect multiple types of mixed coal samples.Based on the analysis of feature extraction of spectral data and analysis of matrix effect dilemmas in LIBS coal quality detection,a cluster-regression model is proposed.This model is based on the characteristic spectral line information in the spectral data measured by LIBS,uses the clustering method to distinguish coal samples with different bases,and different cluster clusters as new ones in the training set.The concept of category which divides the mixed sample set into a subset of multiple categories to train different regression model parameters.For samples to be tested with unknown coal type information,the model can classify them into categories based on the clustering result information of the training set.The corresponding regression model parameters are used to calculate the predicted value of the coal sample ash value index.The cluster regression model can solve the problem of traditional quantitative analysis methods on the unknown sample prediction problem and improve the regression model to a certain extent in complex samples.On the basis,this paper designs an algorithm that can automatically set the clustering parameters for the clustering process in the clustering regression model,which solves the problem of dependence on sample labeling information involved in the manual setting of the density clustering parameters and greatly reduces the difficult for the operator to use it in production environment.Furthermore,the 337 numbered mixed type sample set obtained by Xuzhou coal quality inspection centerand the LIBS test data of 60 coal samples from local coal mine in Shanxi Province are compared.The validity and performance level of the model have been verified.On multiple types of mixed sample sets,the average deviation of the sample ash value calculated by the cluster regression model using the automatic parameter finding algorithm is down to 0.78%,which is much better than the average of the multiple linear regression model.The bias is reduced by 1.52%.A very significant accuracy improvement was achieved,proving the superiority of the cluster regression model in processing mixed sample sets.The proposed clustering-regression model enables LIBS coal quality detection to overcome the limitations of traditional quantitative analysis methods in unknown sample analysis and mixed sample set analysis,and improves the accuracy of existing regression models.It provides the basis and possibilities to achieve high accuracy and stability and LIBS online detection systems as well as other applications.