Accuracy improvement of quantitative analysis of calorific value of coal by combining support vector machine and partial least square methods in laser-induced breakdown spectroscopy

Abstract

Laser-induced breakdown spectroscopy (LIBS) is a potential technology for online coal property analysis, but successful quantitative measurement of calorific value using LIBS suffers from relatively low accuracy caused by the matrix effect. To solve this problem, the support vector machine (SVM) and the partial least square (PLS) were combined to increase the measurement accuracy of calorific value in this study. The combination model utilized SVM to classify coal samples into two groups according to their volatile matter contents to reduce the matrix effect, and then applied PLS to establish calibration models for each sample group respectively. The proposed model was applied to the measurement of calorific values of 53 coal samples, showing that the proposed model could greatly increase accuracy of the measurement of calorific values. Compared with the traditional PLS method, the coefficient of determination (R 2 ) was improved from 0.93 to 0.97, the root-mean-square error of prediction was reduced from 1.68 MJ kg −1 to 1.08 MJ kg −1 , and the average relative error was decreased from 6.7% to 3.93%, showing an overall improvement.

A data selection method for matrix effects and uncertainty reduction for laser-induced breakdown spectroscopy

“…LIBS has been widely used in many scenarios, such as coal analysis [2][3][4], food quality monitoring [5,6], biological tissues [7], the nuclear industry [8,9] and measurement of rare earth elements [10,11]. However, the quantitative performance of LIBS is severely limited by high signal uncertainty and matrix effects.…”

Improving quantitative analysis of cement elements in laser-induced breakdown spectroscopy through combining matrix matching with regression

“…To deal with the above issue, a practical approach is matrix matching, wherein samples are segregated into different groups based on different types of sample matrices. Current research primarily employs methodologies, such as traditional machine learning classiers, [22][23][24] and adaptive subset matching (ASM), 25 to group the samples. An individual model is then built for each group of samples, and multiple models are used to improve the accuracy of quantication.…”

“…PLSR models were then established for different groups of spectra, yielding improved accuracy in the determination of volatile matter, ash content and caloric value. 22,23 Similar studies further consider the similarity of sample matrix properties and the optimization of multiple regression models. 24,25 In the aforementioned studies, the sample division process is based on supervised learning, which requires prior or auxiliary category information.…”

Composition analysis of ceramic raw materials using laser-induced breakdown spectroscopy and autoencoder neural network

“…20 The chemical element composition and content of ceramic raw materials are quite different, which leads to serious matrix effects that will reduce the accuracy of the quantitative analysis model. 21 Multivariate analysis methods, such as SVM, 22,23 PLS, [24][25][26] and limit learning machine 31 have been widely used in LIBS qualitative and quantitative analysis. In recent years, inspired by the success of articial neural networks (ANN) in the eld of articial intelligence, 32,33 more and more researchers tried to apply ANN to quantitative analysis including Back Propagation Neural Network (BPNN) [27][28][29] and convolutional neural network (CNN).…”