Identification of gas mixtures by a distributed support vector machine network and wavelet decomposition from temperature modulated semiconductor gas sensor

Semiconductor gas sensors are wildly applied in gas identification and temperature-modulated dynamic test method is frequently used to improve sensors’ selectivity and stability. This paper introduces a new strategy combing wavelet decomposition and a distributed support vector machine network to identify gas mixtures. In a previous study concerning identifying gases of a few types, we had demonstrated that wavelet decomposition outperforms FFT in feature extraction from responses of thermally modulated semiconductor gas sensors. Here, we extend this method of recognizing gas mixtures of a large type and generalize the rules of selecting the character variables from decomposed wavelet coefficients as character variables. Besides, a distributed support vector machine network is introduced to discriminate the character variables into gas patterns instead of traditional BP neural network. Experiment results show that the strategy proposed here can perform accurate discrimination between hydrogen (H2), carbon monoxide (CO), ethane (C2H4) and their mixtures (total seven patterns) over a range of 50–1000ppm, by using only one commercial metal oxide semiconductor gas sensor, and the discrimination result has advantages in restraining gas concentration variation, sensor's drift and environmental changes. Experimental result also suggests that a distributed support vector machine outperforms a single support vector machine and a BP neural network in performing gas mixture identification.