MOS传感器阵列的二元混合气体检测方法研究

Abstract:In view of the low accuracy of the existing gas detection methods, a novel detection method for binary hybrid gas is proposed. Adopting multichannel nonlinear feature extraction ability, the kernel principal component analysis (KPCA) algorithm is used to extract the binary gas mixtures under different compositions. And the K-nearest neighbor classifier is utlized to achieve the target gas identification. Multivariate relevance vector machine (MVRVM) is used to measure the composition of binary mixture gas by taking advantage of its multivariate nonlinear regression performance. An experimental system is designed including CO gas and CH₄ gas to obtain experimental samples, and the proposed binary mixed gas detection method is verified. The experimental results illustrate that, compared to gas identification methods based on principal componet analysis (PCA) and independent component analysis (ICA), the gas recognition accuracy of proposed method improves 5.83% and 14.16%, respectively, and reaches to 98.33%. Compared to gas estimation methods based on the single RVM and least squaret support vector regression (LS-SVR), the proposed method based on MVRVM effectively improves the accuracy, and the average relative errors of CO and CH₄ concentration estimation are only 5.58% and 5.38% respectively.