Research On On-line Measurement Of Gas Components In Gas-Solid Two-phase Flow

Two-phase flow in the coal mine includes flammable and explosive gas and coal particles,brings a great threat to workers'health and lives.Therefore,eliminating the interference of coal particles is important for high efficient,rapid and real-time on-line detection of mine gas,also it is significant for the safety of mine workers and can avoid gas explosion accidents.In this thesis,simulating the real environment of mine,and an experiment system was set up to collect the gas spectra of methane?CH₄?,ethane?C₂H₆?,propane?C₃H₈?,butane(C₄H₁₀)and carbon dioxide?CO₂?.The original spectral data is denoised by wavelet transform,and then the feature extraction and dimension reduction of the data after noise reduction are carried out by two kinds of feature extraction methods:peak extraction method?PEM?and principal component analysis?PCA?.Then,a support vector machine?SVM?and its improved model and extreme learning machine?ELM?and its improved model were established to quantitatively analyze the mixed gas in the gas-solid two-phase flow in the mine,Taking the methane?CH₄?as an example,the prediction accuracy of the SVM model was analyzed by different parameter optimization methods.The PSO algorithm was used to optimize the penalty parameter c and the kernel function parameter g of the SVM algorithm.PEM-SVM model and improved PCA-SVM model was established based on the PSO algorithm.The prediction accuracy of the improved PEM-SVM model is high when the three kinds of low concentration gas of C₂H₆,C₃H₈ and C₄H₁₀ are predicted.The accuracy of the improved PCA-SVM model is high when the high concentration of CH₄ and CO₂ are predicted.Integrating the advantages of the two models,the PEM-PCA-SVM integrated model was built,And in order to compare with PEM-PCA-SVM model,establishing PEM-PCA-ELM model,which integrated the advantages of PEM-D-ELM model and PCA-D-ELM model.The PEM-PCA-ELM model was tested and verified by using the test set samples,the results show that:The PEM-PCA-ELM model is found to be faster,but the prediction accuracy of PEM-PCA-SVM model is not as good as PEM-PCA-SVM model.