| The problem of metal oxide gas sensor detecting combustible gas used in the mine has been systematically studied in this paper. The detecting mechanism of metal oxide gas sensor is elaborated. Gas recognition model is established to recognize species and concentration of gas sensor response signal. According to requirements of gas sensor using in mine, the system experimental platform is built. With time-frequency analysis methods-short time Fourier transform, wavelet transform method and Hilbert-Huang transform, frequency domain features of gas signal are extracted and spectrums are analyzed and compared. A detailed study of frequency analysis method of gas signal based on atomic decomposition is conducted. Improvement method of match pursuit based on the Gram-Schmidt orthogonalization and adjusting weighting coefficients is proposed to improve resolution and sparsity and reduce processing time. Feature extraction methods based on atomic decomposition parameter feature and time-frequency matrix singular value are presented. With pattern recognition methods of BP neural network, particle swarm optimization with BP hybrid neural network and support vector machine, the simulation experiments of gas recognition are carried out to achieve preset recognition target and meet measurement requirements in mine. |
PDF Full Text Request