| Mine fire and gas explosion have been severely threatening the safety production of coal mine industry in China, and restricted the development of coal mining enterprises. Therefore, the mine gas should be detected accurately and quickly. Presently the infrared spectroscopy has the advantages of fast analysis speed,simple operation and easiness to realize multi-component gases online measurement.The defect of infrared spectroscopy is its low detection precision, but this can be compensated with algorithm. Fourier transform infrared spectroscopy was used in this paper, and two quantitative analysis models were built to reach the accurate and fast detection of mine gas. The main contents and conclusions are as follows:(1) The experimental system based on the infrared spectroscopy for analyzing the composition of natural gas is established, and the detailed experimental procedures are developed to collect the infrared spectral data. The spectral data of126 groups of one-component gas and 110 groups of multi-component gas with five mine gas including methane, ethane, propane, n-butane and carbon dioxide are collected.(2) The spectral data of 236 groups of mine gas are divided into 186 groups for calibration sets and 50 groups for validation sets. The quantitative analysis model of mine gas based on kernel partial least-squares(KPLS) was built. And1 1(2200 ~ 2400) cm(2800 ~ 3200) cm--? was selected as the spectral analysis regions. Through the analysis and comparison, the optimization of preprocessing was adopted. And the quantitative analysis model based on KPLS was established and verified by the validation sets. The experiment results are as follows: the maximum relative error of each kind of gas is no more than 8%, the average relative error is no more than 3%, the maximum relative error of each sample gas is 3.26%, and the average relative error of 50 groups of validation sets is 1.25%. The results show that the quantitative analysis model based on KPLS can reach the mine gas detection, but has slightly larger deviation and is unstable.(3) The quantitative analysis model of mine gas based on improved support vector machine(SVM) was built. Principal component analysis(PCA) was used toreduce the dimensionality of the infrared spectral data. 3 eigenvalues were extracted as input, which helped to improve convergence speed and reduce calculation time.Particle swarm optimization(PSO) and genetic algorithm(GA) were used to optimize parameters of support vector machine(SVM) method respectively, and PSO was adopted for its better optimization effect over GA. The experiment results are as follows: the maximum relative error of each kind of gas is no more than 5%, the average relative error is no more than 2%, the maximum relative error of each sample gas is 2.19%, and the average relative error of 50 groups of validation sets is0.68%. The results show that the quantitative analysis model based on improved SVM is better than normal SVM and KPLS, which has better accurate and stable detection results. |
